diff --git a/benchmark/README.md b/benchmark/README.md
index 68fd5ae2e9..d929d8c908 100644
--- a/benchmark/README.md
+++ b/benchmark/README.md
@@ -1,6 +1,11 @@
-# Sigmoid Kernel Benchmarks
+# Darktable Mojo Benchmarks
 
-This directory contains benchmarks and parity checks for the darktable `sigmoid` module kernels, comparing the traditional **OpenCL (C)** implementation against the **Mojo GPU** implementation.
+This directory contains performance benchmarks and validation tools for darktable module kernels ported to Mojo.
+
+## Subprojects
+
+- **[Sigmoid](./sigmoid)**: Benchmarks and parity checks for the sigmoid module.
+- **[Blurs](./blurs)**: Benchmarks for the blurs convolution module.
 
 ## Prerequisites
 
@@ -10,49 +15,8 @@ This directory contains benchmarks and parity checks for the darktable `sigmoid`
 
 ## Environment Setup
 
-Initialize the environment using Pixi:
+Initialize the environment using Pixi in the root `benchmark/` directory:
 
 ```bash
 pixi install
-# or enter the shell
-pixi shell
-```
-
-## Running Benchmarks
-
-### 1. OpenCL Baseline (C)
-The C benchmark measures the performance of the original OpenCL kernels.
-
-```bash
-make run
 ```
-
-### 2. Mojo GPU Benchmark
-The Mojo benchmark measures the performance of the ported kernels using Mojo's GPU abstraction.
-
-```bash
-pixi run mojo sigmoid_benchmark_gpu.mojo
-```
-
-## Parity Validation
-
-To ensure the Mojo implementation produces numerically identical results to the OpenCL baseline, run the parity check script:
-
-```bash
-pixi run python validate_parity.py
-```
-
-This script:
-1. Compiles and runs the OpenCL parity check (`parity_check.c`).
-2. Runs the Mojo GPU benchmark.
-3. Compares sampled pixels across both implementations (RGB Ratio and Per-Channel modes).
-4. Reports "PASS" if the results match within a tolerance of $10^{-6}$.
-
-## Files
-
-- `benchmark_sigmoid.c`: Main C-based OpenCL benchmark.
-- `sigmoid_benchmark_gpu.mojo`: Mojo implementation and benchmark.
-- `parity_check.c`: Minimal OpenCL runner for numerical validation.
-- `validate_parity.py`: Automated comparison tool.
-- `Makefile`: Build instructions for C/OpenCL binaries.
-- `pixi.toml`: Project dependencies (Mojo, Python, etc.).
diff --git a/benchmark/blurs/Makefile b/benchmark/blurs/Makefile
new file mode 100644
index 0000000000..e16dc6b570
--- /dev/null
+++ b/benchmark/blurs/Makefile
@@ -0,0 +1,18 @@
+CC=gcc
+CFLAGS=-O3 -Wall
+LDFLAGS=-lOpenCL -lm
+
+TARGET=benchmark_blurs
+SRC=benchmark_blurs.c
+
+all: $(TARGET)
+	
+.PHONY: force
+$(TARGET): $(SRC) force
+	$(CC) $(CFLAGS) $(SRC) -o $(TARGET) $(LDFLAGS)
+
+clean:
+	rm -f $(TARGET)
+
+run: all
+	./$(TARGET)
diff --git a/benchmark/blurs/README.md b/benchmark/blurs/README.md
new file mode 100644
index 0000000000..efae830150
--- /dev/null
+++ b/benchmark/blurs/README.md
@@ -0,0 +1,30 @@
+# Blurs Kernel Benchmarks
+
+This directory contains benchmarks for the darktable `blurs` module kernels, comparing the **OpenCL (C)** implementation against the **Mojo GPU (Tiled)** implementation.
+
+## Running Benchmarks
+
+### 1. OpenCL Baseline (C)
+```bash
+make run
+```
+
+### 2. Mojo GPU Benchmark
+Run from the `benchmark/blurs` directory. Use `-I` to point to the `mojo/` source root.
+
+```bash
+mojo -I /path/to/darktable/mojo blurs_benchmark_gpu.mojo
+```
+
+Or if using pixi:
+```bash
+pixi run mojo -I /path/to/darktable/mojo blurs_benchmark_gpu.mojo
+```
+
+From the darktable project root:
+```bash
+cd benchmark/blurs && mojo -I ../../mojo blurs_benchmark_gpu.mojo
+```
+
+## Performance Notes
+The Mojo implementation uses a tiled approach with SIMD vectorization to ensure coalesced memory access on the GPU. This is designed to be significantly faster than a naive implementation.
diff --git a/benchmark/blurs/benchmark_blurs b/benchmark/blurs/benchmark_blurs
new file mode 100755
index 0000000000..498cf22a4f
Binary files /dev/null and b/benchmark/blurs/benchmark_blurs differ
diff --git a/benchmark/blurs/benchmark_blurs.c b/benchmark/blurs/benchmark_blurs.c
new file mode 100644
index 0000000000..38a80a7406
--- /dev/null
+++ b/benchmark/blurs/benchmark_blurs.c
@@ -0,0 +1,180 @@
+#define CL_TARGET_OPENCL_VERSION 120
+#include <CL/cl.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <time.h>
+#include <math.h>
+
+#define CHECK_CL(cmd) \
+    { \
+        cl_int _cl_err = cmd; \
+        if (_cl_err != CL_SUCCESS) { \
+            fprintf(stderr, "OpenCL error %d at %s:%d\n", _cl_err, __FILE__, __LINE__); \
+            exit(1); \
+        } \
+    }
+
+char* read_file(const char* filename) {
+    FILE* f = fopen(filename, "rb");
+    if (!f) return NULL;
+    fseek(f, 0, SEEK_END);
+    long size = ftell(f);
+    fseek(f, 0, SEEK_SET);
+    char* buf = (char*)malloc(size + 1);
+    if (!buf) {
+        fclose(f);
+        return NULL;
+    }
+    if (fread(buf, 1, size, f) != (size_t)size) {
+        free(buf);
+        fclose(f);
+        return NULL;
+    }
+    buf[size] = '\0';
+    fclose(f);
+    return buf;
+}
+
+static double run_benchmark(cl_context context, cl_command_queue queue, cl_program program,
+                            cl_mem d_in, int width, int height, int radius) {
+    int k_width = 2 * radius + 1;
+    int iterations = 100;
+    if (radius >= 12) iterations = 30;
+    else if (radius >= 5) iterations = 80;
+
+    // Create kernel + output image (reusable)
+    cl_image_format format = { CL_RGBA, CL_FLOAT };
+    cl_image_desc desc = { CL_MEM_OBJECT_IMAGE2D, width, height, 0, 0, 0, 0, 0, 0, {NULL} };
+    cl_int err;
+    cl_mem d_out = clCreateImage(context, CL_MEM_WRITE_ONLY, &format, &desc, NULL, &err);
+    CHECK_CL(err);
+
+    // Create kernel image for this radius
+    size_t k_data_size = (size_t)k_width * k_width * sizeof(float);
+    float* h_kern = (float*)malloc(k_data_size);
+    for (int i = 0; i < k_width * k_width; i++)
+        h_kern[i] = 1.0f / (k_width * k_width);
+
+    cl_image_format kern_format = { CL_R, CL_FLOAT };
+    cl_image_desc kern_desc = { CL_MEM_OBJECT_IMAGE2D, k_width, k_width, 0, 0, 0, 0, 0, 0, {NULL} };
+    cl_mem d_kern = clCreateImage(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
+                                  &kern_format, &kern_desc, h_kern, &err);
+    CHECK_CL(err);
+
+    cl_kernel kernel = clCreateKernel(program, "convolve", &err);
+    CHECK_CL(err);
+
+    CHECK_CL(clSetKernelArg(kernel, 0, sizeof(cl_mem), &d_in));
+    CHECK_CL(clSetKernelArg(kernel, 1, sizeof(cl_mem), &d_kern));
+    CHECK_CL(clSetKernelArg(kernel, 2, sizeof(cl_mem), &d_out));
+    CHECK_CL(clSetKernelArg(kernel, 3, sizeof(int), &width));
+    CHECK_CL(clSetKernelArg(kernel, 4, sizeof(int), &height));
+    CHECK_CL(clSetKernelArg(kernel, 5, sizeof(int), &radius));
+
+    printf("Benchmarking OpenCL convolve (Radius: %d, %dx%d, %d iters)...\n", radius, width, height, iterations);
+    fflush(stdout);
+
+    cl_event event;
+    double total_time = 0;
+    for (int i = 0; i < iterations; i++) {
+        size_t global_work_size[2] = { (size_t)width, (size_t)height };
+        CHECK_CL(clEnqueueNDRangeKernel(queue, kernel, 2, NULL, global_work_size, NULL, 0, NULL, &event));
+        clWaitForEvents(1, &event);
+        cl_ulong start, end;
+        clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_START, sizeof(start), &start, NULL);
+        clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_END, sizeof(end), &end, NULL);
+        total_time += (double)(end - start) / 1000000.0;
+        clReleaseEvent(event);
+    }
+    double avg = total_time / iterations;
+    printf("  Average Time: %.4f ms\n", avg);
+
+    clReleaseMemObject(d_kern);
+    clReleaseMemObject(d_out);
+    clReleaseKernel(kernel);
+    free(h_kern);
+    return avg;
+}
+
+int main() {
+    printf("Starting C benchmark...\n");
+    fflush(stdout);
+
+    cl_int err;
+    cl_platform_id platform;
+    cl_device_id device;
+    cl_context context;
+    cl_command_queue queue;
+    int width = 6016;
+    int height = 4016;
+    size_t img_size = (size_t)width * height * 4 * sizeof(float);
+
+    printf("Allocating memory for %dx%d image...\n", width, height);
+    fflush(stdout);
+    float* h_data = (float*)malloc(img_size);
+    if (!h_data) { fprintf(stderr, "Failed to allocate h_data\n"); return 1; }
+
+    printf("Initializing host data (%zu bytes)...\n", img_size);
+    fflush(stdout);
+    for (size_t i = 0; i < (size_t)width * height * 4; i++) h_data[i] = 0.5f;
+    printf("Host data initialized.\n");
+    fflush(stdout);
+
+    // OpenCL Initialization
+    printf("Initializing OpenCL...\n");
+    fflush(stdout);
+    cl_uint num_platforms;
+    err = clGetPlatformIDs(0, NULL, &num_platforms);
+    if (err != CL_SUCCESS || num_platforms == 0) {
+        fprintf(stderr, "No OpenCL platforms found\n");
+        return 1;
+    }
+    CHECK_CL(clGetPlatformIDs(1, &platform, NULL));
+    CHECK_CL(clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL));
+
+    char device_name[128];
+    clGetDeviceInfo(device, CL_DEVICE_NAME, sizeof(device_name), device_name, NULL);
+    printf("Using device: %s\n", device_name);
+
+    context = clCreateContext(NULL, 1, &device, NULL, NULL, &err);
+    CHECK_CL(err);
+    queue = clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &err);
+    CHECK_CL(err);
+
+    // Load kernel source
+    const char* kernel_path = "../../data/kernels/blurs.cl";
+    char* source = read_file(kernel_path);
+    if (!source) { fprintf(stderr, "Failed to load kernel\n"); return 1; }
+
+    cl_program program = clCreateProgramWithSource(context, 1, (const char**)&source, NULL, &err);
+    CHECK_CL(err);
+    const char* options = "-I ../../data/kernels/";
+    err = clBuildProgram(program, 1, &device, options, NULL, NULL);
+    if (err != CL_SUCCESS) {
+        char log[16384];
+        clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, sizeof(log), log, NULL);
+        fprintf(stderr, "Build error:\n%s\n", log);
+        return 1;
+    }
+
+    // Create input image (reused across radii)
+    cl_image_format format = { CL_RGBA, CL_FLOAT };
+    cl_image_desc desc = { CL_MEM_OBJECT_IMAGE2D, width, height, 0, 0, 0, 0, 0, 0, {NULL} };
+    cl_mem d_in = clCreateImage(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
+                                &format, &desc, h_data, &err);
+    CHECK_CL(err);
+
+    int radii[] = {3, 8, 15};
+    int num_radii = sizeof(radii) / sizeof(radii[0]);
+    for (int r = 0; r < num_radii; r++) {
+        run_benchmark(context, queue, program, d_in, width, height, radii[r]);
+    }
+
+    free(source); free(h_data);
+    clReleaseMemObject(d_in);
+    clReleaseProgram(program);
+    clReleaseCommandQueue(queue);
+    clReleaseContext(context);
+    return 0;
+}
diff --git a/benchmark/blurs/benchmark_results.md b/benchmark/blurs/benchmark_results.md
new file mode 100644
index 0000000000..2571b2f9ef
--- /dev/null
+++ b/benchmark/blurs/benchmark_results.md
@@ -0,0 +1,59 @@
+# GPU Convolution Benchmark Results
+
+An execution-time comparison on a massive 24.1 Megapixel ($6016 \times 4016$, 4-channel float32) image, comparing:
+1. **OpenCL C-Benchmark** (`benchmark_blurs.c` using C/OpenCL host)
+2. **Mojo GPU Naive Baseline** (replicating the global memory `UnsafePointer` implementation from `iop/blurs` in Mojo)
+3. **Mojo GPU Tiled Convolution** (using a single compiled kernel with dynamic runtime `radius` parameter, zero `UnsafePointer` usage)
+
+---
+
+## Benchmark Configuration
+
+- **Image Resolution**: $6016 \times 4016$ (4 channels, RGBA)
+- **Data Type**: `Float32`
+- **Kernel Size**: $31 \times 31$ Box Blur (`RADIUS = 15`)
+- **Kernel Initialization**: Normalized Box Blur (each element = $1 / 961$)
+- **Target GPU**: AMD HIP/OpenCL GPU (`gfx1201`)
+- **LDS Buffer Allocation Size**: Sized to support any dynamic runtime `radius` up to $25$
+
+---
+
+## Performance Summary
+
+| Implementation | Average Latency (ms) | Rel. Performance |
+| :--- | :--- | :--- |
+| **Mojo GPU Tiled (`TileTensor`)** | **38.5 ms** | **1.65x** (Fastest) 🏆 |
+| **Mojo GPU Naive Baseline** | **63 ms** | **1.00x** |
+| **OpenCL convolve (C-Benchmark)** | **69 ms** | **0.91x** |
+
+---
+
+## Technical Analysis & Key Discoveries
+
+### 1. Zero-Specialization Dynamic Runtime Radius
+We transitioned the GPU kernel from compile-time specialization of the radius to a pure, user-selected runtime parameter:
+```mojo
+def convolve_gpu_kernel[
+    InLayout: TensorLayout,
+    OutLayout: TensorLayout,
+    max_radius: Int,       # Sizing upper bound for LDS buffer allocation
+](
+    in_t: TileTensor[DTYPE, InLayout, MutAnyOrigin],
+    out_t: TileTensor[DTYPE, OutLayout, MutAnyOrigin],
+    k_val: Float32,        # Precomputed box-blur weight (1 / kern_size)
+    radius: Int,           # Dynamic runtime radius parameter
+    ...
+)
+```
+- **LDS Allocation Sized by `MAX_RADIUS = 25`**: A compile-time constant `MAX_RADIUS` sizes the shared memory array. This single compilation supports *any* runtime user-selected radius $r \le 25$ without generating multiple specializations.
+- **Division-Free 2D Strided Loader**: Instead of mapping a flat thread index using slow runtime modulo and division operations (which is highly detrimental when radius is dynamic), we developed an elegant 2D strided loader:
+  ```mojo
+  for ly in range(ty, actual_halo_h, TILE_H):
+      for lx in range(tx, actual_halo_w, TILE_W):
+          ...
+          sh_tile[c, ly, lx] = val
+  ```
+  This eliminates all division/modulo instructions from the loading stage and preserves perfectly coalesced, bank-conflict-free shared memory writes.
+
+### 2. Double-Occupancy via Alpha-Elimination
+By omitting the Alpha channel from the shared tile layout (`row_major[RGB, HALO_H, HALO_W]`), we reduced the LDS footprint from **45.6 KB** to **34.2 KB**. This allows **2 active blocks per CU** on gfx1201's 64 KB LDS limit, doubling latency hiding and keeping execution throughput at an optimal level.
diff --git a/benchmark/blurs/check_cl.c b/benchmark/blurs/check_cl.c
new file mode 100644
index 0000000000..e31e82bda3
--- /dev/null
+++ b/benchmark/blurs/check_cl.c
@@ -0,0 +1,14 @@
+#include <CL/cl.h>
+#include <stdio.h>
+
+int main() {
+    printf("Checking OpenCL platforms...\n");
+    cl_uint num_platforms;
+    cl_int err = clGetPlatformIDs(0, NULL, &num_platforms);
+    if (err != CL_SUCCESS) {
+        printf("clGetPlatformIDs failed with %d\n", err);
+        return 1;
+    }
+    printf("Found %u platforms.\n", num_platforms);
+    return 0;
+}
diff --git a/benchmark/Makefile b/benchmark/sigmoid/Makefile
similarity index 100%
rename from benchmark/Makefile
rename to benchmark/sigmoid/Makefile
diff --git a/benchmark/sigmoid/README.md b/benchmark/sigmoid/README.md
new file mode 100644
index 0000000000..68fd5ae2e9
--- /dev/null
+++ b/benchmark/sigmoid/README.md
@@ -0,0 +1,58 @@
+# Sigmoid Kernel Benchmarks
+
+This directory contains benchmarks and parity checks for the darktable `sigmoid` module kernels, comparing the traditional **OpenCL (C)** implementation against the **Mojo GPU** implementation.
+
+## Prerequisites
+
+- **Mojo**: Required for running `.mojo` benchmarks.
+- **OpenCL**: Required for hardware acceleration on GPU (headers and libraries).
+- **Pixi**: Used for environment and dependency management.
+
+## Environment Setup
+
+Initialize the environment using Pixi:
+
+```bash
+pixi install
+# or enter the shell
+pixi shell
+```
+
+## Running Benchmarks
+
+### 1. OpenCL Baseline (C)
+The C benchmark measures the performance of the original OpenCL kernels.
+
+```bash
+make run
+```
+
+### 2. Mojo GPU Benchmark
+The Mojo benchmark measures the performance of the ported kernels using Mojo's GPU abstraction.
+
+```bash
+pixi run mojo sigmoid_benchmark_gpu.mojo
+```
+
+## Parity Validation
+
+To ensure the Mojo implementation produces numerically identical results to the OpenCL baseline, run the parity check script:
+
+```bash
+pixi run python validate_parity.py
+```
+
+This script:
+1. Compiles and runs the OpenCL parity check (`parity_check.c`).
+2. Runs the Mojo GPU benchmark.
+3. Compares sampled pixels across both implementations (RGB Ratio and Per-Channel modes).
+4. Reports "PASS" if the results match within a tolerance of $10^{-6}$.
+
+## Files
+
+- `benchmark_sigmoid.c`: Main C-based OpenCL benchmark.
+- `sigmoid_benchmark_gpu.mojo`: Mojo implementation and benchmark.
+- `parity_check.c`: Minimal OpenCL runner for numerical validation.
+- `validate_parity.py`: Automated comparison tool.
+- `Makefile`: Build instructions for C/OpenCL binaries.
+- `pixi.toml`: Project dependencies (Mojo, Python, etc.).
diff --git a/benchmark/benchmark_sigmoid.c b/benchmark/sigmoid/benchmark_sigmoid.c
similarity index 85%
rename from benchmark/benchmark_sigmoid.c
rename to benchmark/sigmoid/benchmark_sigmoid.c
index 4b8f32ad6b..2e55b13d1f 100644
--- a/benchmark/benchmark_sigmoid.c
+++ b/benchmark/sigmoid/benchmark_sigmoid.c
@@ -148,14 +148,10 @@ int main() {
     // Create command queue with profiling enabled
     cl_queue_properties props[] = { CL_QUEUE_PROPERTIES, CL_QUEUE_PROFILING_ENABLE, 0 };
     queue = clCreateCommandQueueWithProperties(context, device, props, &err);
-    if (err != CL_SUCCESS) {
-        // Fallback for older OpenCL versions if needed, but we targeting 2.2 above
-        queue = clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &err);
-    }
     CHECK_CL(err);
 
     // Load kernel source
-    const char* kernel_path = "../data/kernels/sigmoid.cl";
+    const char* kernel_path = "../../data/kernels/sigmoid.cl";
     printf("Reading kernel source from %s...\n", kernel_path);
     char* source = read_file(kernel_path);
     if (!source) {
@@ -168,7 +164,7 @@ int main() {
     CHECK_CL(err);
     
     // Need to point to the directory containing common.h and colorspace.h
-    const char* options = "-I ../data/kernels/";
+    const char* options = "-I ../../data/kernels/";
     err = clBuildProgram(program, 1, &device, options, NULL, NULL);
     if (err != CL_SUCCESS) {
         char build_log[16384];
@@ -183,9 +179,6 @@ int main() {
     cl_kernel kernel_rgb_ratio = clCreateKernel(program, "sigmoid_loglogistic_rgb_ratio", &err);
     CHECK_CL(err);
 
-    CHECK_CL(clGetPlatformIDs(1, &platform, &num_platforms));
-    CHECK_CL(clGetDeviceIDs(platform, CL_DEVICE_TYPE_GPU, 1, &device, NULL));
-
     cl_image_format format = { CL_RGBA, CL_FLOAT };
     cl_image_desc desc = { CL_MEM_OBJECT_IMAGE2D, width, height, 0, 0, 0, 0, 0, 0, {0} };
     
@@ -233,10 +226,11 @@ int main() {
     CHECK_CL(clSetKernelArg(kernel_per_channel, 11, sizeof(cl_mem), &d_mat2));
     CHECK_CL(clSetKernelArg(kernel_per_channel, 12, sizeof(cl_mem), &d_mat3));
 
-    int warmup_iters = 100;
-    int iterations = 1000;
+    int warmup_iters = 20;
+    int iterations = 200;
+    int batch_size = 20;
+    int num_batches = iterations / batch_size;
     cl_event event;
-    double total_time = 0;
 
     printf("  Warmup (%d iterations)...\n", warmup_iters);
     for (int i = 0; i < warmup_iters; i++) {
@@ -245,7 +239,8 @@ int main() {
     }
     clFinish(queue);
 
-    printf("  Benchmarking (%d iterations)...\n", iterations);
+    printf("  Benchmarking (%d iterations, batch_size=%d)...\n", iterations, batch_size);
+    double *times = (double*)malloc(iterations * sizeof(double));
     for (int i = 0; i < iterations; i++) {
         size_t global_work_size[2] = { width, height };
         CHECK_CL(clEnqueueNDRangeKernel(queue, kernel_per_channel, 2, NULL, global_work_size, NULL, 0, NULL, &event));
@@ -254,10 +249,29 @@ int main() {
         cl_ulong start, end;
         clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_START, sizeof(start), &start, NULL);
         clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_END, sizeof(end), &end, NULL);
-        total_time += (double)(end - start) / 1000000.0; // ns to ms
+        times[i] = (double)(end - start) / 1000000.0; // ns to ms
         clReleaseEvent(event);
     }
-    printf("  Average Time: %.4f ms\n", total_time / iterations);
+
+    double *batch_means = (double*)malloc(num_batches * sizeof(double));
+    double overall_sum = 0;
+    for (int b = 0; b < num_batches; b++) {
+        double batch_sum = 0;
+        for (int j = 0; j < batch_size; j++)
+            batch_sum += times[b * batch_size + j];
+        batch_means[b] = batch_sum / batch_size;
+        overall_sum += batch_means[b];
+    }
+    double mean = overall_sum / num_batches;
+    double sum_sq = 0;
+    for (int b = 0; b < num_batches; b++) {
+        double d = batch_means[b] - mean;
+        sum_sq += d * d;
+    }
+    double std = sqrt(sum_sq / num_batches);
+    free(batch_means);
+    free(times);
+    printf("  Mean: %.4f ms, Std: %.4f ms\n", mean, std);
 
     // Save result of the last iteration to a file
     float* h_out = (float*)malloc(img_size);
@@ -289,8 +303,8 @@ int main() {
     }
     clFinish(queue);
 
-    total_time = 0;
-    printf("  Benchmarking (%d iterations)...\n", iterations);
+    printf("  Benchmarking (%d iterations, batch_size=%d)...\n", iterations, batch_size);
+    times = (double*)malloc(iterations * sizeof(double));
     for (int i = 0; i < iterations; i++) {
         size_t global_work_size[2] = { width, height };
         CHECK_CL(clEnqueueNDRangeKernel(queue, kernel_rgb_ratio, 2, NULL, global_work_size, NULL, 0, NULL, &event));
@@ -299,10 +313,29 @@ int main() {
         cl_ulong start, end;
         clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_START, sizeof(start), &start, NULL);
         clGetEventProfilingInfo(event, CL_PROFILING_COMMAND_END, sizeof(end), &end, NULL);
-        total_time += (double)(end - start) / 1000000.0; // ns to ms
+        times[i] = (double)(end - start) / 1000000.0; // ns to ms
         clReleaseEvent(event);
     }
-    printf("  Average Time: %.4f ms\n", total_time / iterations);
+
+    batch_means = (double*)malloc(num_batches * sizeof(double));
+    overall_sum = 0;
+    for (int b = 0; b < num_batches; b++) {
+        double batch_sum = 0;
+        for (int j = 0; j < batch_size; j++)
+            batch_sum += times[b * batch_size + j];
+        batch_means[b] = batch_sum / batch_size;
+        overall_sum += batch_means[b];
+    }
+    mean = overall_sum / num_batches;
+    sum_sq = 0;
+    for (int b = 0; b < num_batches; b++) {
+        double d = batch_means[b] - mean;
+        sum_sq += d * d;
+    }
+    std = sqrt(sum_sq / num_batches);
+    free(batch_means);
+    free(times);
+    printf("  Mean: %.4f ms, Std: %.4f ms\n", mean, std);
 
     // Save result of the last iteration to a file
     h_out = (float*)malloc(img_size);
diff --git a/benchmark/sigmoid/opencl_input.jpg b/benchmark/sigmoid/opencl_input.jpg
new file mode 100644
index 0000000000..e5d0773daf
Binary files /dev/null and b/benchmark/sigmoid/opencl_input.jpg differ
diff --git a/benchmark/sigmoid/opencl_output_per_channel.jpg b/benchmark/sigmoid/opencl_output_per_channel.jpg
new file mode 100644
index 0000000000..76d7b4902d
Binary files /dev/null and b/benchmark/sigmoid/opencl_output_per_channel.jpg differ
diff --git a/benchmark/sigmoid/opencl_output_rgb_ratio.jpg b/benchmark/sigmoid/opencl_output_rgb_ratio.jpg
new file mode 100644
index 0000000000..08176fb2ca
Binary files /dev/null and b/benchmark/sigmoid/opencl_output_rgb_ratio.jpg differ
diff --git a/benchmark/parity_check.c b/benchmark/sigmoid/parity_check.c
similarity index 100%
rename from benchmark/parity_check.c
rename to benchmark/sigmoid/parity_check.c
diff --git a/benchmark/sigmoid/sigmoid_benchmark_gpu.mojo b/benchmark/sigmoid/sigmoid_benchmark_gpu.mojo
new file mode 100644
index 0000000000..45059945ff
--- /dev/null
+++ b/benchmark/sigmoid/sigmoid_benchmark_gpu.mojo
@@ -0,0 +1,246 @@
+from std.gpu.host import DeviceContext, DeviceBuffer, HostBuffer
+from std.utils import Index
+from std.math import sqrt
+from std.benchmark import Bench, BenchConfig, Bencher, BenchId
+from layout import Layout, LayoutTensor, UNKNOWN_VALUE
+from layout.runtime_layout import RuntimeLayout, IndexList
+from std.memory.unsafe_pointer import UnsafePointer
+from iop.sigmoid.kernels import (
+    apply_sigmoid_rgb_ratio,
+    apply_sigmoid_per_channel,
+)
+from iop.sigmoid.lib import (
+    _launch_rgb_ratio_gpu,
+    _launch_per_channel_gpu,
+)
+
+comptime WIDTH = 6016
+comptime HEIGHT = 4016
+comptime CHANNELS = 4
+comptime IMAGE_LAYOUT = Layout.row_major(UNKNOWN_VALUE, UNKNOWN_VALUE, CHANNELS)
+comptime DTYPE = DType.float32
+
+
+def main() raises:
+    var total_floats = HEIGHT * WIDTH * CHANNELS
+    var ctx = DeviceContext()
+    print("Using GPU API:", ctx.api())
+
+    var input_buffer_host = ctx.enqueue_create_host_buffer[DTYPE](total_floats)
+    var input_buffer_device = ctx.enqueue_create_buffer[DTYPE](total_floats)
+    var output_buffer_device = ctx.enqueue_create_buffer[DTYPE](total_floats)
+
+    var rt = RuntimeLayout[IMAGE_LAYOUT].row_major(
+        IndexList[3](HEIGHT, WIDTH, CHANNELS)
+    )
+
+    var input_image_host = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
+        UnsafePointer[Float32, MutAnyOrigin](
+            unsafe_from_address=Int(input_buffer_host.unsafe_ptr())
+        ),
+        rt,
+    )
+    var input_image_device = LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin](
+        UnsafePointer[Float32, ImmutAnyOrigin](
+            unsafe_from_address=Int(input_buffer_device.unsafe_ptr())
+        ),
+        rt,
+    )
+    var output_image_device = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
+        UnsafePointer[Float32, MutAnyOrigin](
+            unsafe_from_address=Int(output_buffer_device.unsafe_ptr())
+        ),
+        rt,
+    )
+
+    for y in range(HEIGHT):
+        var h = Float32(y) / (HEIGHT - 1) * 6.0
+        var segment = Int(h)
+        var f1 = h - Float32(segment)
+        var pr: Float32
+        var pg: Float32
+        var pb: Float32
+        if segment == 0:
+            pr = 1.0; pg = f1; pb = 0.0
+        elif segment == 1:
+            pr = 1.0 - f1; pg = 1.0; pb = 0.0
+        elif segment == 2:
+            pr = 0.0; pg = 1.0; pb = f1
+        elif segment == 3:
+            pr = 0.0; pg = 1.0 - f1; pb = 1.0
+        elif segment == 4:
+            pr = f1; pg = 0.0; pb = 1.0
+        elif segment == 5:
+            pr = 1.0; pg = 0.0; pb = 1.0 - f1
+        else:
+            pr = 1.0; pg = 0.0; pb = 0.0
+
+        for x in range(WIDTH):
+            var r: Float32
+            var g: Float32
+            var b: Float32
+            var mid_x = Float32(WIDTH) / 2.0
+            if Float32(x) < mid_x:
+                var t = Float32(x) / mid_x
+                r = pr * t; g = pg * t; b = pb * t
+            else:
+                var t = (Float32(x) - mid_x) / (Float32(WIDTH) - 1.0 - mid_x)
+                r = pr * (1.0 - t) + t
+                g = pg * (1.0 - t) + t
+                b = pb * (1.0 - t) + t
+            input_image_host.store[width=4](
+                Index(y, x, 0), SIMD[DType.float32, 4](r, g, b, 1.0)
+            )
+
+    ctx.enqueue_copy(input_buffer_device, input_buffer_host)
+
+    var white_target = Float32(1.0)
+    var black_target = Float32(0.000152)
+    var paper_exp = Float32(0.5)
+    var film_fog = Float32(0.0)
+    var contrast_power = Float32(2.5)
+    var skew_power = Float32(1.0)
+    var hue_preservation = Float32(1.0)
+
+    var identity = SIMD[DType.float32, 16](0)
+    identity[0] = 1; identity[5] = 1; identity[10] = 1; identity[15] = 1
+
+    _launch_rgb_ratio_gpu(
+        ctx,
+        input_image_device,
+        output_image_device,
+        white_target,
+        black_target,
+        paper_exp,
+        film_fog,
+        contrast_power,
+        skew_power,
+        WIDTH,
+        HEIGHT,
+        WIDTH * HEIGHT,
+    )
+    ctx.synchronize()
+
+    var output_buffer_host = ctx.enqueue_create_host_buffer[DTYPE](total_floats)
+    ctx.enqueue_copy(output_buffer_host, output_buffer_device)
+    ctx.synchronize()
+    var output_image_host = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
+        UnsafePointer[Float32, MutAnyOrigin](
+            unsafe_from_address=Int(output_buffer_host.unsafe_ptr())
+        ),
+        rt,
+    )
+
+    for i in range(1, 4):
+        var t = Float32(i) * 0.2
+        var x = Int(t * Float32(WIDTH))
+        var y = Int(t * Float32(HEIGHT))
+        var res = output_image_host.load[width=4](Index(y, x, 0))
+        print(
+            "RGB Ratio -", Int(t * 100), "% pixel: [",
+            res[0], res[1], res[2], res[3], "]",
+        )
+
+    _launch_per_channel_gpu(
+        ctx,
+        input_image_device,
+        output_image_device,
+        white_target,
+        paper_exp,
+        film_fog,
+        contrast_power,
+        skew_power,
+        hue_preservation,
+        identity,
+        identity,
+        identity,
+        WIDTH,
+        HEIGHT,
+        WIDTH * HEIGHT,
+    )
+    ctx.synchronize()
+
+    ctx.enqueue_copy(output_buffer_host, output_buffer_device)
+    ctx.synchronize()
+
+    for i in range(1, 4):
+        var t = Float32(i) * 0.2
+        var x = Int(t * Float32(WIDTH))
+        var y = Int(t * Float32(HEIGHT))
+        var res = output_image_host.load[width=4](Index(y, x, 0))
+        print(
+            "Per Channel -", Int(t * 100), "% pixel: [",
+            res[0], res[1], res[2], res[3], "]",
+        )
+
+    var bench = Bench(BenchConfig(
+        max_iters=200, num_warmup_iters=20, max_batch_size=20,
+    ))
+
+    @parameter
+    def bench_rgb(mut b: Bencher) raises:
+        @parameter
+        def run_rgb(ctx: DeviceContext) raises:
+            _launch_rgb_ratio_gpu(
+                ctx,
+                input_image_device,
+                output_image_device,
+                white_target,
+                black_target,
+                paper_exp,
+                film_fog,
+                contrast_power,
+                skew_power,
+                WIDTH,
+                HEIGHT,
+                WIDTH * HEIGHT,
+            )
+
+        b.iter_custom[run_rgb](ctx)
+        ctx.synchronize()
+
+    @parameter
+    def bench_per(mut b: Bencher) raises:
+        @parameter
+        def run_per(ctx: DeviceContext) raises:
+            _launch_per_channel_gpu(
+                ctx,
+                input_image_device,
+                output_image_device,
+                white_target,
+                paper_exp,
+                film_fog,
+                contrast_power,
+                skew_power,
+                hue_preservation,
+                identity,
+                identity,
+                identity,
+                WIDTH,
+                HEIGHT,
+                WIDTH * HEIGHT,
+            )
+
+        b.iter_custom[run_per](ctx)
+        ctx.synchronize()
+
+    bench.bench_function[bench_rgb](
+        BenchId("Mojo-Sigmoid-GPU-RGB-Ratio-V2"),
+    )
+    bench.bench_function[bench_per](
+        BenchId("Mojo-Sigmoid-GPU-Per-Channel-V2"),
+    )
+    print(bench)
+
+    for idx in range(len(bench.info_vec)):
+        ref info = bench.info_vec[idx]
+        var overall_mean = info.result.mean("ms")
+        var sum_sq = 0.0
+        var v_n = Float64(len(info.result.runs))
+        for b in range(len(info.result.runs)):
+            var batch_mean = info.result.runs[b].mean("ms")
+            var d = batch_mean - overall_mean
+            sum_sq += d * d
+        var variance = sum_sq / v_n
+        var std = sqrt(variance)
+        print(info.name, "- std:", std, "ms")
diff --git a/benchmark/stb_image_write.h b/benchmark/sigmoid/stb_image_write.h
similarity index 100%
rename from benchmark/stb_image_write.h
rename to benchmark/sigmoid/stb_image_write.h
diff --git a/benchmark/validate_parity.py b/benchmark/sigmoid/validate_parity.py
similarity index 100%
rename from benchmark/validate_parity.py
rename to benchmark/sigmoid/validate_parity.py
diff --git a/benchmark/sigmoid_benchmark_gpu.mojo b/benchmark/sigmoid_benchmark_gpu.mojo
deleted file mode 100644
index 0321f52346..0000000000
--- a/benchmark/sigmoid_benchmark_gpu.mojo
+++ /dev/null
@@ -1,601 +0,0 @@
-from gpu import block_dim, block_idx, thread_idx
-from gpu.host import DeviceContext, DeviceBuffer, Dim, HostBuffer
-from layout import Layout, LayoutTensor
-from utils import Index, IndexList
-from math import ceildiv, isnan, sqrt, pow, max, min
-from benchmark import Bench, BenchConfig, Bencher, BenchId
-from algorithm.functional import elementwise
-
-# Configuration
-comptime WIDTH = 6016
-comptime HEIGHT = 4016
-comptime CHANNELS = 4
-comptime IMAGE_LAYOUT = Layout.row_major(HEIGHT, WIDTH, CHANNELS)
-comptime DTYPE = DType.float32
-
-
-# Removed Pixel struct to use 3D LayoutTensor directly as requested
-
-
-@always_inline
-fn generalized_loglogistic_sigmoid_scalar(
-    value: Float32,
-    magnitude: Float32,
-    paper_exp: Float32,
-    film_fog: Float32,
-    film_power: Float32,
-    paper_power: Float32,
-) -> Float32:
-    var clamped_value = max(value, Float32(0.0))
-    var film_response = pow(film_fog + clamped_value, film_power)
-    var paper_response = magnitude * pow(
-        film_response / (paper_exp + film_response), paper_power
-    )
-    if isnan(paper_response):
-        return magnitude
-    return paper_response
-
-
-@always_inline
-fn apply_sigmoid_rgb_ratio(
-    in_r: Float32,
-    in_g: Float32,
-    in_b: Float32,
-    in_a: Float32,
-    white_target: Float32,
-    black_target: Float32,
-    paper_exp: Float32,
-    film_fog: Float32,
-    film_power: Float32,
-    paper_power: Float32,
-) -> SIMD[DType.float32, 4]:
-    # Desaturate negative values
-    var avg = max((in_r + in_g + in_b) / 3.0, Float32(0.0))
-    var min_v = min(min(in_r, in_g), in_b)
-    var sat = Float32(1.0)
-    if min_v < 0.0:
-        sat = -avg / (min_v - avg)
-
-    var p_r = avg + sat * (in_r - avg)
-    var p_g = avg + sat * (in_g - avg)
-    var p_b = avg + sat * (in_b - avg)
-
-    var luma = (p_r + p_g + p_b) / 3.0
-    var mapped_luma = generalized_loglogistic_sigmoid_scalar(
-        luma, white_target, paper_exp, film_fog, film_power, paper_power
-    )
-
-    if luma > 1e-9:
-        var scale = mapped_luma / luma
-        p_r *= scale
-        p_g *= scale
-        p_b *= scale
-    else:
-        p_r = mapped_luma
-        p_g = mapped_luma
-        p_b = mapped_luma
-
-    var p_min = min(min(p_r, p_g), p_b)
-    var p_max = max(max(p_r, p_g), p_b)
-    var eps = Float32(1e-6)
-    var d_white = (white_target - mapped_luma) / (p_max - mapped_luma + eps)
-    var d_black = (black_target - mapped_luma) / (p_min - mapped_luma - eps)
-    var db_vs_chroma = min(d_white, d_black)
-    var cvm_border = (mapped_luma - p_min) / (mapped_luma + eps)
-    var p_chr_adj = 1.0 / (cvm_border * db_vs_chroma + eps)
-    var h_chr = (
-        2.0 * cvm_border / (1.0 - cvm_border * cvm_border + eps)
-    ) * p_chr_adj
-    var h_z = sqrt(h_chr * h_chr + 1.0)
-    var chroma_f = h_chr / (1.0 + h_z) * db_vs_chroma
-
-    return SIMD[DType.float32, 4](
-        mapped_luma + chroma_f * (p_r - mapped_luma),
-        mapped_luma + chroma_f * (p_g - mapped_luma),
-        mapped_luma + chroma_f * (p_b - mapped_luma),
-        in_a,
-    )
-
-
-@always_inline
-fn apply_sigmoid_per_channel(
-    in_r: Float32,
-    in_g: Float32,
-    in_b: Float32,
-    in_a: Float32,
-    white_target: Float32,
-    paper_exp: Float32,
-    film_fog: Float32,
-    contrast_power: Float32,
-    skew_power: Float32,
-    hue_preservation: Float32,
-    pipe_to_base: SIMD[DType.float32, 16],
-    base_to_rendering: SIMD[DType.float32, 16],
-    rendering_to_pipe: SIMD[DType.float32, 16],
-) -> SIMD[DType.float32, 4]:
-    # 1. Transform to base space
-    var i_r = (
-        pipe_to_base[0] * in_r + pipe_to_base[1] * in_g + pipe_to_base[2] * in_b
-    )
-    var i_g = (
-        pipe_to_base[4] * in_r + pipe_to_base[5] * in_g + pipe_to_base[6] * in_b
-    )
-    var i_b = (
-        pipe_to_base[8] * in_r
-        + pipe_to_base[9] * in_g
-        + pipe_to_base[10] * in_b
-    )
-
-    # 2. Desaturate negative
-    var avg = max((i_r + i_g + i_b) / 3.0, Float32(0.0))
-    var min_v = min(min(i_r, i_g), i_b)
-    var sat = Float32(1.0)
-    if min_v < 0.0:
-        sat = -avg / (min_v - avg)
-    i_r = avg + sat * (i_r - avg)
-    i_g = avg + sat * (i_g - avg)
-    i_b = avg + sat * (i_b - avg)
-
-    # 3. Transform to rendering space
-    var r_r = (
-        base_to_rendering[0] * i_r
-        + base_to_rendering[1] * i_g
-        + base_to_rendering[2] * i_b
-    )
-    var r_g = (
-        base_to_rendering[4] * i_r
-        + base_to_rendering[5] * i_g
-        + base_to_rendering[6] * i_b
-    )
-    var r_b = (
-        base_to_rendering[8] * i_r
-        + base_to_rendering[9] * i_g
-        + base_to_rendering[10] * i_b
-    )
-
-    # 4. Per-channel sigmoid curves
-    var pc_r = generalized_loglogistic_sigmoid_scalar(
-        r_r, white_target, paper_exp, film_fog, contrast_power, skew_power
-    )
-    var pc_g = generalized_loglogistic_sigmoid_scalar(
-        r_g, white_target, paper_exp, film_fog, contrast_power, skew_power
-    )
-    var pc_b = generalized_loglogistic_sigmoid_scalar(
-        r_b, white_target, paper_exp, film_fog, contrast_power, skew_power
-    )
-
-    # 5. Preserve hue & energy
-    var p_min: Float32
-    var p_mid: Float32
-    var p_max: Float32
-    var pc_min: Float32
-    var pc_mid: Float32
-    var pc_max: Float32
-
-    if r_r >= r_g:
-        if r_g >= r_b:  # R G B
-            p_max = r_r
-            p_mid = r_g
-            p_min = r_b
-            pc_max = pc_r
-            pc_mid = pc_g
-            pc_min = pc_b
-        elif r_b >= r_r:  # B R G
-            p_max = r_b
-            p_mid = r_r
-            p_min = r_g
-            pc_max = pc_b
-            pc_mid = pc_r
-            pc_min = pc_g
-        else:  # R B G
-            p_max = r_r
-            p_mid = r_b
-            p_min = r_g
-            pc_max = pc_r
-            pc_mid = pc_b
-            pc_min = pc_g
-    else:
-        if r_r >= r_b:  # G R B
-            p_max = r_g
-            p_mid = r_r
-            p_min = r_b
-            pc_max = pc_g
-            pc_mid = pc_r
-            pc_min = pc_b
-        elif r_b >= r_g:  # B G R
-            p_max = r_b
-            p_mid = r_g
-            p_min = r_r
-            pc_max = pc_b
-            pc_mid = pc_g
-            pc_min = pc_r
-        else:  # G B R
-            p_max = r_g
-            p_mid = r_b
-            p_min = r_r
-            pc_max = pc_g
-            pc_mid = pc_b
-            pc_min = pc_r
-
-    var chroma = p_max - p_min
-    var midscale = Float32(0.0)
-    if chroma != 0.0:
-        midscale = (p_mid - p_min) / chroma
-
-    var f_hc = pc_min + (pc_max - pc_min) * midscale
-    var n_mid = (1.0 - hue_preservation) * pc_mid + hue_preservation * f_hc
-
-    var blend = 2.0 * p_min / (p_min + p_mid + 1e-9)
-    var target = blend * (pc_r + pc_g + pc_b) + (1.0 - blend) * (
-        pc_min + n_mid + pc_max
-    )
-
-    var res_min: Float32
-    var res_mid: Float32
-    var res_max: Float32
-    if n_mid <= pc_mid:
-        res_mid = (
-            (1.0 - hue_preservation) * pc_mid
-            + hue_preservation
-            * (midscale * pc_max + (1.0 - midscale) * (target - pc_max))
-        ) / (1.0 + hue_preservation * (1.0 - midscale))
-        res_min = target - pc_max - res_mid
-        res_max = pc_max
-    else:
-        res_mid = (
-            (1.0 - hue_preservation) * pc_mid
-            + hue_preservation
-            * (pc_min * (1.0 - midscale) + midscale * (target - pc_min))
-        ) / (1.0 + hue_preservation * midscale)
-        res_min = pc_min
-        res_max = target - pc_min - res_mid
-
-    var res_r: Float32
-    var res_g: Float32
-    var res_b: Float32
-    if r_r >= r_g:
-        if r_g >= r_b:  # R G B
-            res_r = res_max
-            res_g = res_mid
-            res_b = res_min
-        elif r_b >= r_r:  # B R G
-            res_b = res_max
-            res_r = res_mid
-            res_g = res_min
-        else:  # R B G
-            res_r = res_max
-            res_b = res_mid
-            res_g = res_min
-    else:
-        if r_r >= r_b:  # G R B
-            res_g = res_max
-            res_r = res_mid
-            res_b = res_min
-        elif r_b >= r_g:  # B G R
-            res_b = res_max
-            res_g = res_mid
-            res_r = res_min
-        else:  # G B R
-            res_g = res_max
-            res_b = res_mid
-            res_r = res_min
-
-    # 6. Transform to pipe space
-    var out_r = (
-        rendering_to_pipe[0] * res_r
-        + rendering_to_pipe[1] * res_g
-        + rendering_to_pipe[2] * res_b
-    )
-    var out_g = (
-        rendering_to_pipe[4] * res_r
-        + rendering_to_pipe[5] * res_g
-        + rendering_to_pipe[6] * res_b
-    )
-    var out_b = (
-        rendering_to_pipe[8] * res_r
-        + rendering_to_pipe[9] * res_g
-        + rendering_to_pipe[10] * res_b
-    )
-
-    return SIMD[DType.float32, 4](out_r, out_g, out_b, in_a)
-
-
-fn run_sigmoid_rgb_ratio(
-    ctx: DeviceContext,
-    output: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
-    input: LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin],
-    white_target: Float32,
-    black_target: Float32,
-    paper_exp: Float32,
-    film_fog: Float32,
-    film_power: Float32,
-    paper_power: Float32,
-    num_pixels: Int,
-) raises:
-    @parameter
-    @always_inline
-    fn rgb_ratio_closure[
-        width: Int, rank: Int, alignment: Int
-    ](indices: IndexList[rank]) capturing -> None:
-        var px_idx = indices[0]
-        var y = px_idx // WIDTH
-        var x = px_idx % WIDTH
-        var pix = input.load[width=4](Index(y, x, 0))
-        var res = apply_sigmoid_rgb_ratio(
-            pix[0],
-            pix[1],
-            pix[2],
-            pix[3],
-            white_target,
-            black_target,
-            paper_exp,
-            film_fog,
-            film_power,
-            paper_power,
-        )
-        output.store[width=4](Index(y, x, 0), res)
-
-    elementwise[rgb_ratio_closure, 1, target="gpu"](num_pixels, ctx)
-
-
-fn run_sigmoid_per_channel(
-    ctx: DeviceContext,
-    output: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
-    input: LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin],
-    white_target: Float32,
-    paper_exp: Float32,
-    film_fog: Float32,
-    contrast_power: Float32,
-    skew_power: Float32,
-    hue_preservation: Float32,
-    pipe_to_base: SIMD[DType.float32, 16],
-    base_to_rendering: SIMD[DType.float32, 16],
-    rendering_to_pipe: SIMD[DType.float32, 16],
-    num_pixels: Int,
-) raises:
-    @parameter
-    @always_inline
-    fn per_channel_closure[
-        width: Int, rank: Int, alignment: Int
-    ](indices: IndexList[rank]) capturing -> None:
-        var px_idx = indices[0]
-        var y = px_idx // WIDTH
-        var x = px_idx % WIDTH
-        var pix = input.load[width=4](Index(y, x, 0))
-        var res = apply_sigmoid_per_channel(
-            pix[0],
-            pix[1],
-            pix[2],
-            pix[3],
-            white_target,
-            paper_exp,
-            film_fog,
-            contrast_power,
-            skew_power,
-            hue_preservation,
-            pipe_to_base,
-            base_to_rendering,
-            rendering_to_pipe,
-        )
-        output.store[width=4](Index(y, x, 0), res)
-
-    elementwise[per_channel_closure, 1, target="gpu"](num_pixels, ctx)
-
-
-fn main() raises:
-    var total_floats = HEIGHT * WIDTH * CHANNELS
-    var ctx = DeviceContext()
-    print("Using GPU API:", ctx.api())
-
-    var input_buffer_host = ctx.enqueue_create_host_buffer[DTYPE](total_floats)
-    var input_buffer_device = ctx.enqueue_create_buffer[DTYPE](total_floats)
-    var output_buffer_device = ctx.enqueue_create_buffer[DTYPE](total_floats)
-
-    var input_image_host = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
-        input_buffer_host
-    )
-    var input_image_device = LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin](
-        input_buffer_device
-    )
-    var output_image_device = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
-        output_buffer_device
-    )
-
-    for y in range(HEIGHT):
-        var h = Float32(y) / (HEIGHT - 1) * 6.0
-        var segment = Int(h)
-        var f1 = h - Float32(segment)
-        var pr: Float32
-        var pg: Float32
-        var pb: Float32
-        if segment == 0:
-            pr = 1.0
-            pg = f1
-            pb = 0.0
-        elif segment == 1:
-            pr = 1.0 - f1
-            pg = 1.0
-            pb = 0.0
-        elif segment == 2:
-            pr = 0.0
-            pg = 1.0
-            pb = f1
-        elif segment == 3:
-            pr = 0.0
-            pg = 1.0 - f1
-            pb = 1.0
-        elif segment == 4:
-            pr = f1
-            pg = 0.0
-            pb = 1.0
-        elif segment == 5:
-            pr = 1.0
-            pg = 0.0
-            pb = 1.0 - f1
-        else:
-            pr = 1.0
-            pg = 0.0
-            pb = 0.0
-
-        for x in range(WIDTH):
-            var r: Float32
-            var g: Float32
-            var b: Float32
-            var mid_x = Float32(WIDTH) / 2.0
-            if Float32(x) < mid_x:
-                var t = Float32(x) / mid_x
-                r = pr * t
-                g = pg * t
-                b = pb * t
-            else:
-                var t = (Float32(x) - mid_x) / (Float32(WIDTH) - 1.0 - mid_x)
-                r = pr * (1.0 - t) + t
-                g = pg * (1.0 - t) + t
-                b = pb * (1.0 - t) + t
-            input_image_host.store[width=4](
-                Index(y, x, 0), SIMD[DType.float32, 4](r, g, b, 1.0)
-            )
-
-    ctx.enqueue_copy(input_buffer_device, input_buffer_host)
-
-    var white_target = Float32(1.0)
-    var black_target = Float32(0.000152)
-    var paper_exp = Float32(0.5)
-    var film_fog = Float32(0.0)
-    var contrast_power = Float32(2.5)
-    var skew_power = Float32(1.0)
-    var hue_preservation = Float32(1.0)
-
-    var identity = SIMD[DType.float32, 16](0)
-    identity[0] = 1
-    identity[5] = 1
-    identity[10] = 1
-    identity[15] = 1
-
-    # Runs
-    run_sigmoid_rgb_ratio(
-        ctx,
-        output_image_device,
-        input_image_device,
-        white_target,
-        black_target,
-        paper_exp,
-        film_fog,
-        contrast_power,
-        skew_power,
-        WIDTH * HEIGHT,
-    )
-    ctx.synchronize()
-
-    var output_buffer_host = ctx.enqueue_create_host_buffer[DTYPE](total_floats)
-    ctx.enqueue_copy(output_buffer_host, output_buffer_device)
-    ctx.synchronize()
-    var output_image_host = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
-        output_buffer_host
-    )
-
-    for i in range(1, 4):
-        var t = Float32(i) * 0.2
-        var x = Int(t * Float32(WIDTH))
-        var y = Int(t * Float32(HEIGHT))
-        var res = output_image_host.load[width=4](Index(y, x, 0))
-        print(
-            "RGB Ratio -",
-            Int(t * 100),
-            "% pixel: [",
-            res[0],
-            res[1],
-            res[2],
-            res[3],
-            "]",
-        )
-
-    run_sigmoid_per_channel(
-        ctx,
-        output_image_device,
-        input_image_device,
-        white_target,
-        paper_exp,
-        film_fog,
-        contrast_power,
-        skew_power,
-        hue_preservation,
-        identity,
-        identity,
-        identity,
-        WIDTH * HEIGHT,
-    )
-    ctx.synchronize()
-
-    ctx.enqueue_copy(output_buffer_host, output_buffer_device)
-    ctx.synchronize()
-
-    for i in range(1, 4):
-        var t = Float32(i) * 0.2
-        var x = Int(t * Float32(WIDTH))
-        var y = Int(t * Float32(HEIGHT))
-        var res = output_image_host.load[width=4](Index(y, x, 0))
-        print(
-            "Per Channel -",
-            Int(t * 100),
-            "% pixel: [",
-            res[0],
-            res[1],
-            res[2],
-            res[3],
-            "]",
-        )
-
-    # Benchmarking
-    var bench = Bench(BenchConfig(max_iters=1000, num_warmup_iters=100))
-
-    @parameter
-    fn bench_rgb(mut b: Bencher) raises:
-        @parameter
-        fn run(ctx: DeviceContext) raises:
-            run_sigmoid_rgb_ratio(
-                ctx,
-                output_image_device,
-                input_image_device,
-                white_target,
-                black_target,
-                paper_exp,
-                film_fog,
-                contrast_power,
-                skew_power,
-                WIDTH * HEIGHT,
-            )
-
-        b.iter_custom[run](ctx)
-        ctx.synchronize()
-
-    @parameter
-    fn bench_per(mut b: Bencher) raises:
-        @parameter
-        fn run(ctx: DeviceContext) raises:
-            run_sigmoid_per_channel(
-                ctx,
-                output_image_device,
-                input_image_device,
-                white_target,
-                paper_exp,
-                film_fog,
-                contrast_power,
-                skew_power,
-                hue_preservation,
-                identity,
-                identity,
-                identity,
-                WIDTH * HEIGHT,
-            )
-
-        b.iter_custom[run](ctx)
-        ctx.synchronize()
-
-    bench.bench_function[bench_rgb](
-        BenchId("Mojo-Sigmoid-GPU-RGB-Ratio-V2"), fixed_iterations=1000
-    )
-    bench.bench_function[bench_per](
-        BenchId("Mojo-Sigmoid-GPU-Per-Channel-V2"), fixed_iterations=1000
-    )
-    print(bench)
diff --git a/mojo/.gitattributes b/mojo/.gitattributes
new file mode 100644
index 0000000000..997504b465
--- /dev/null
+++ b/mojo/.gitattributes
@@ -0,0 +1,2 @@
+# SCM syntax highlighting & preventing 3-way merges
+pixi.lock merge=binary linguist-language=YAML linguist-generated=true -diff
diff --git a/mojo/.gitignore b/mojo/.gitignore
new file mode 100644
index 0000000000..ae849e65b8
--- /dev/null
+++ b/mojo/.gitignore
@@ -0,0 +1,3 @@
+# pixi environments
+.pixi/*
+!.pixi/config.toml
diff --git a/mojo/Makefile b/mojo/Makefile
new file mode 100644
index 0000000000..c8648a298b
--- /dev/null
+++ b/mojo/Makefile
@@ -0,0 +1,17 @@
+MOJO = pixi run mojo
+PLUGIN_DIR = /usr/lib/darktable/plugins
+
+.PHONY: all build install bench clean
+
+all: build
+
+build: libsigmoid_mojo.so
+
+libsigmoid_mojo.so: iop/sigmoid/lib.mojo iop/sigmoid/kernels.mojo
+	$(MOJO) build -I . iop/sigmoid/lib.mojo --emit shared-lib -o libsigmoid_mojo.so
+
+install: libsigmoid_mojo.so
+	sudo cp libsigmoid_mojo.so $(PLUGIN_DIR)/
+
+clean:
+	rm -f libsigmoid_mojo.so
diff --git a/mojo/build_sigmoid_iop.sh b/mojo/build_sigmoid_iop.sh
new file mode 100755
index 0000000000..0c06170049
--- /dev/null
+++ b/mojo/build_sigmoid_iop.sh
@@ -0,0 +1,51 @@
+#!/bin/bash
+# Build libsigmoid.so from the darktable source tree.
+# This mimics what CMake does (run introspection, then compile + link).
+
+set -e
+
+SRC=/home/mc/code/darktable
+BUILD=/tmp/sigmoid_build
+DEST=$BUILD/libsigmoid.so
+
+mkdir -p "$BUILD"
+
+# Step 1: Run introspection (generates introspection_sigmoid.c)
+echo "=== Step 1: Introspection ==="
+perl "$SRC/tools/introspection/parser.pl" \
+    "$SRC/src/" \
+    "$SRC/src/iop/sigmoid.c" \
+    "$BUILD/introspection_sigmoid.c"
+echo "  -> $BUILD/introspection_sigmoid.c ($(wc -l < "$BUILD/introspection_sigmoid.c") lines)"
+
+# Step 2: Collect flags
+GTK_FLAGS=$(pkg-config --cflags gtk+-3.0 glib-2.0 librsvg-2.0 lcms2 json-glib-1.0)
+DT_INCLUDE="-I$SRC/src -I$SRC/src/iop -I$SRC/build/bin -I/usr/lib/darktable"
+DEFINES="-DHAVE_CONFIG_H -DHAVE_OPENCL -D_GNU_SOURCE -include common/module_api.h -include iop/iop_api.h"
+CFLAGS="-O3 -march=native -fPIC -fvisibility=hidden -fopenmp $GTK_FLAGS $DT_INCLUDE $DEFINES"
+LDFLAGS="-L/usr/lib/darktable -ldarktable -lm -lgomp -Wl,-rpath,/usr/lib/darktable"
+
+echo ""
+echo "=== Step 2: Compile ==="
+gcc $CFLAGS \
+    -Wno-unused-function \
+    -Wno-deprecated-declarations \
+    -c "$BUILD/introspection_sigmoid.c" \
+    -o "$BUILD/introspection_sigmoid.o" 2>&1
+echo "  -> $BUILD/introspection_sigmoid.o"
+
+echo ""
+echo "=== Step 3: Link ==="
+gcc -shared -fPIC -fopenmp \
+    "$BUILD/introspection_sigmoid.o" \
+    $LDFLAGS \
+    -o "$DEST"
+echo "  -> $DEST"
+
+echo ""
+echo "=== Step 4: Verify ==="
+nm -D "$DEST" | grep " T " | awk '{print $3}'
+
+echo ""
+echo "Build complete: $DEST"
+echo "Size: $(du -sh "$DEST" | cut -f1)"
diff --git a/mojo/iop/__init__.mojo b/mojo/iop/__init__.mojo
new file mode 100644
index 0000000000..0c393fc1b4
--- /dev/null
+++ b/mojo/iop/__init__.mojo
@@ -0,0 +1 @@
+# iop package
diff --git a/mojo/iop/blurs/__init__.mojo b/mojo/iop/blurs/__init__.mojo
new file mode 100644
index 0000000000..8abcdf557a
--- /dev/null
+++ b/mojo/iop/blurs/__init__.mojo
@@ -0,0 +1 @@
+# iop.blurs package
diff --git a/mojo/iop/blurs/kernels.mojo b/mojo/iop/blurs/kernels.mojo
new file mode 100644
index 0000000000..0898c38d54
--- /dev/null
+++ b/mojo/iop/blurs/kernels.mojo
@@ -0,0 +1,59 @@
+from layout import Layout, LayoutTensor, UNKNOWN_VALUE
+from std.math import clamp
+from std.utils import Index
+from std.memory.unsafe_pointer import UnsafePointer
+
+comptime CHANNELS = 4
+comptime IMAGE_LAYOUT = Layout.row_major(UNKNOWN_VALUE, UNKNOWN_VALUE, CHANNELS)
+comptime DTYPE = DType.float32
+
+
+@always_inline
+def apply_convolve_vector[W: Int](
+    x: Int,
+    y: Int,
+    width: Int,
+    height: Int,
+    radius: Int,
+    in_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin],
+    kern_ptr: UnsafePointer[Float32, ImmutAnyOrigin],
+) -> SIMD[DType.float32, W * 4]:
+    var acc = SIMD[DType.float32, W * 4](0)
+
+    for l in range(-radius, radius + 1):
+        var ii = clamp(y + l, 0, height - 1)
+        for m in range(-radius, radius + 1):
+            var ik = l + radius
+            var jk = m + radius
+            var k_width = 2 * radius + 1
+            var k = kern_ptr[ik * k_width + jk]
+
+            for i in range(W):
+                var jj = clamp(x + i + m, 0, width - 1)
+                var pix = in_t.load[width=4](Index(ii, jj, 0))
+
+                acc[i * 4] += k * pix[0]
+                acc[i * 4 + 1] += k * pix[1]
+                acc[i * 4 + 2] += k * pix[2]
+                acc[i * 4 + 3] += k * pix[3]
+
+    for i in range(W):
+        var curr_x = x + i
+        if curr_x < width:
+            var orig = in_t.load[width=4](Index(y, curr_x, 0))
+            acc[i * 4 + 3] = orig[3]
+
+    return acc
+
+
+@always_inline
+def apply_convolve(
+    x: Int,
+    y: Int,
+    width: Int,
+    height: Int,
+    radius: Int,
+    in_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin],
+    kern_ptr: UnsafePointer[Float32, ImmutAnyOrigin],
+) -> SIMD[DType.float32, 4]:
+    return apply_convolve_vector[1](x, y, width, height, radius, in_t, kern_ptr)
diff --git a/mojo/iop/blurs/lib.mojo b/mojo/iop/blurs/lib.mojo
new file mode 100644
index 0000000000..c8e966add0
--- /dev/null
+++ b/mojo/iop/blurs/lib.mojo
@@ -0,0 +1,190 @@
+from std.gpu.host import DeviceContext
+from layout import Layout, LayoutTensor, UNKNOWN_VALUE
+from layout.runtime_layout import RuntimeLayout, IndexList
+from layout.coord import Coord
+from std.utils import Index
+from std.algorithm.functional import elementwise, vectorize
+from std.gpu.host.compile import get_gpu_target
+from std.memory import alloc
+from std.memory.unsafe_pointer import UnsafePointer
+from std.sys import simd_width_of
+from iop.blurs.kernels import apply_convolve, apply_convolve_vector
+
+comptime CHANNELS = 4
+comptime IMAGE_LAYOUT = Layout.row_major(UNKNOWN_VALUE, UNKNOWN_VALUE, CHANNELS)
+comptime DTYPE = DType.float32
+comptime SIMD_WIDTH = simd_width_of[DTYPE, target=get_gpu_target()]()
+
+
+struct MojoCtx:
+    var use_gpu: Int
+    var dctx_addr: Int
+
+    def __init__(out self, use_gpu: Int, dctx_addr: Int):
+        self.use_gpu = use_gpu
+        self.dctx_addr = dctx_addr
+
+
+def _launch_convolve_gpu(
+    dctx: DeviceContext,
+    in_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin],
+    out_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
+    kern_ptr: UnsafePointer[Float32, ImmutAnyOrigin],
+    width: Int,
+    height: Int,
+    radius: Int,
+    num_pixels: Int,
+) raises:
+    @parameter
+    @always_inline
+    def gpu_kernel[simd_width: Int, alignment: Int](coord: Coord) capturing -> None:
+        var px_idx = Int(coord[0].value())
+        var y = px_idx // width
+        var x = px_idx % width
+        var res = apply_convolve_vector[1](
+            x, y, width, height, radius, in_t, kern_ptr
+        )
+        out_t.store[width=4](Index(y, x, 0), res)
+
+    elementwise[gpu_kernel, 1, target="gpu"](num_pixels, dctx)
+    dctx.synchronize()
+
+
+@export("blurs_mojo_init")
+def blurs_mojo_init(
+    ctx_out: UnsafePointer[Int, MutAnyOrigin], use_gpu: Int32
+) abi("C") -> None:
+    var gpu = use_gpu != 0
+    var dctx_addr = 0
+    if gpu:
+        try:
+            var d_ptr = alloc[DeviceContext](1)
+            d_ptr.unsafe_write(DeviceContext())
+            dctx_addr = Int(d_ptr)
+            print("Mojo Blurs: GPU Context Initialized")
+        except e:
+            print("Mojo Blurs: GPU Init Error:", String(e))
+            gpu = False
+    else:
+        print("Mojo Blurs: CPU Context Initialized")
+    var p = alloc[MojoCtx](1)
+    p[0].use_gpu = 1 if gpu else 0
+    p[0].dctx_addr = dctx_addr
+    ctx_out[0] = Int(p)
+
+
+@export("blurs_mojo_destroy")
+def blurs_mojo_destroy(ctx_addr: Int) abi("C") -> None:
+    var p = UnsafePointer[MojoCtx, MutAnyOrigin](unsafe_from_address=ctx_addr)
+    if p[0].dctx_addr != 0:
+        var dctx_ptr = UnsafePointer[DeviceContext, MutAnyOrigin](
+            unsafe_from_address=p[0].dctx_addr
+        )
+        dctx_ptr.unsafe_deinit_pointee()
+        dctx_ptr.free()
+    p.free()
+
+
+@export("blurs_mojo_convolve")
+def blurs_mojo_convolve(
+    ctx_addr: Int,
+    in_addr: Int,
+    kern_addr: Int,
+    out_addr: Int,
+    width: Int32,
+    height: Int32,
+    radius: Int32,
+) abi("C") -> None:
+    var ctx_p = UnsafePointer[MojoCtx, MutAnyOrigin](
+        unsafe_from_address=ctx_addr
+    )
+    var use_gpu = ctx_p[0].use_gpu != 0
+    var dctx_addr = ctx_p[0].dctx_addr
+
+    var kern_p = UnsafePointer[Float32, MutAnyOrigin](
+        unsafe_from_address=kern_addr
+    )
+    var out_p = UnsafePointer[Float32, MutAnyOrigin](
+        unsafe_from_address=out_addr
+    )
+
+    var h = Int(height)
+    var w = Int(width)
+    var r = Int(radius)
+    var num_pixels = h * w
+    var k_size = (2 * r + 1) * (2 * r + 1)
+
+    var rt = RuntimeLayout[IMAGE_LAYOUT].row_major(
+        IndexList[3](h, w, CHANNELS)
+    )
+
+    if use_gpu and dctx_addr != 0:
+        try:
+            var in_p = UnsafePointer[Float32, MutAnyOrigin](
+                unsafe_from_address=in_addr
+            )
+            var dctx = UnsafePointer[DeviceContext, MutAnyOrigin](
+                unsafe_from_address=dctx_addr
+            )[0]
+            var dev_in = dctx.enqueue_create_buffer[DTYPE](num_pixels * 4)
+            var dev_kern = dctx.enqueue_create_buffer[DTYPE](k_size)
+            var dev_out = dctx.enqueue_create_buffer[DTYPE](num_pixels * 4)
+
+            dctx.enqueue_copy(dev_in, in_p)
+            dctx.enqueue_copy(dev_kern, kern_p)
+
+            var in_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin](
+                UnsafePointer[Float32, ImmutAnyOrigin](
+                    unsafe_from_address=Int(dev_in.unsafe_ptr())
+                ),
+                rt,
+            )
+            var out_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
+                UnsafePointer[Float32, MutAnyOrigin](
+                    unsafe_from_address=Int(dev_out.unsafe_ptr())
+                ),
+                rt,
+            )
+            var kern_ptr_gpu = UnsafePointer[Float32, ImmutAnyOrigin](
+                unsafe_from_address=Int(dev_kern.unsafe_ptr())
+            )
+
+            _launch_convolve_gpu(
+                dctx, in_t, out_t, kern_ptr_gpu, w, h, r, num_pixels,
+            )
+
+            dctx.enqueue_copy(out_p, dev_out)
+            dctx.synchronize()
+        except e:
+            print("GPU Run Error (Convolve):", String(e))
+    else:
+        var in_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin](
+            UnsafePointer[Float32, ImmutAnyOrigin](
+                unsafe_from_address=in_addr
+            ),
+            rt,
+        )
+        var out_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
+            out_p,
+            rt,
+        )
+        var kern_ptr_cpu = UnsafePointer[Float32, ImmutAnyOrigin](
+            unsafe_from_address=kern_addr
+        )
+
+        for y in range(h):
+            var cur_y = y
+            def row_fn[width: Int](x: Int) {mut}:
+                var res = apply_convolve_vector[width](
+                    x, cur_y, w, h, r, in_t, kern_ptr_cpu
+                )
+                for i in range(width):
+                    var pixel = SIMD[DType.float32, 4](
+                        res[i * 4],
+                        res[i * 4 + 1],
+                        res[i * 4 + 2],
+                        res[i * 4 + 3],
+                    )
+                    out_t.store[width=4](Index(cur_y, x + i, 0), pixel)
+
+            vectorize[SIMD_WIDTH](w, row_fn)
diff --git a/mojo/iop/sigmoid/__init__.mojo b/mojo/iop/sigmoid/__init__.mojo
new file mode 100644
index 0000000000..256cd8163f
--- /dev/null
+++ b/mojo/iop/sigmoid/__init__.mojo
@@ -0,0 +1 @@
+# iop.sigmoid package
diff --git a/mojo/iop/sigmoid/kernels.mojo b/mojo/iop/sigmoid/kernels.mojo
new file mode 100644
index 0000000000..8bfeb61285
--- /dev/null
+++ b/mojo/iop/sigmoid/kernels.mojo
@@ -0,0 +1,204 @@
+from std.math import isnan, sqrt, pow, max, min
+
+@always_inline
+def generalized_loglogistic_sigmoid_scalar(
+    value: Float32,
+    magnitude: Float32,
+    paper_exp: Float32,
+    film_fog: Float32,
+    film_power: Float32,
+    paper_power: Float32,
+) -> Float32:
+    var clamped_value = max(value, Float32(0.0))
+    var film_response = pow(film_fog + clamped_value, film_power)
+    var paper_response = magnitude * pow(
+        film_response / (paper_exp + film_response), paper_power
+    )
+    if isnan(paper_response):
+        return magnitude
+    return paper_response
+
+@always_inline
+def apply_sigmoid_rgb_ratio(
+    in_r: Float32,
+    in_g: Float32,
+    in_b: Float32,
+    in_a: Float32,
+    white_target: Float32,
+    black_target: Float32,
+    paper_exp: Float32,
+    film_fog: Float32,
+    film_power: Float32,
+    paper_power: Float32,
+) -> SIMD[DType.float32, 4]:
+    var avg = max((in_r + in_g + in_b) / Float32(3.0), Float32(0.0))
+    var min_v = min(min(in_r, in_g), in_b)
+    var sat = Float32(1.0)
+    if min_v < Float32(0.0):
+        sat = -avg / (min_v - avg)
+
+    var p_r = avg + sat * (in_r - avg)
+    var p_g = avg + sat * (in_g - avg)
+    var p_b = avg + sat * (in_b - avg)
+
+    var luma = (p_r + p_g + p_b) / Float32(3.0)
+    var mapped_luma = generalized_loglogistic_sigmoid_scalar(
+        luma, white_target, paper_exp, film_fog, film_power, paper_power
+    )
+
+    if luma > Float32(1e-9):
+        var scale = mapped_luma / luma
+        p_r *= scale
+        p_g *= scale
+        p_b *= scale
+    else:
+        p_r = mapped_luma
+        p_g = mapped_luma
+        p_b = mapped_luma
+
+    var p_min = min(min(p_r, p_g), p_b)
+    var p_max = max(max(p_r, p_g), p_b)
+    var eps = Float32(1e-6)
+    var d_white = (white_target - mapped_luma) / (p_max - mapped_luma + eps)
+    var d_black = (black_target - mapped_luma) / (p_min - mapped_luma - eps)
+    var db_vs_chroma = min(d_white, d_black)
+    var cvm_border = (mapped_luma - p_min) / (mapped_luma + eps)
+    var p_chr_adj = Float32(1.0) / (cvm_border * db_vs_chroma + eps)
+    var h_chr = (Float32(2.0) * cvm_border / (Float32(1.0) - cvm_border * cvm_border + eps)) * p_chr_adj
+    var h_z = sqrt(h_chr * h_chr + Float32(1.0))
+    var chroma_f = h_chr / (Float32(1.0) + h_z) * db_vs_chroma
+
+    return SIMD[DType.float32, 4](
+        mapped_luma + chroma_f * (p_r - mapped_luma),
+        mapped_luma + chroma_f * (p_g - mapped_luma),
+        mapped_luma + chroma_f * (p_b - mapped_luma),
+        in_a,
+    )
+
+@always_inline
+def apply_sigmoid_per_channel(
+    in_r: Float32,
+    in_g: Float32,
+    in_b: Float32,
+    in_a: Float32,
+    white_target: Float32,
+    paper_exp: Float32,
+    film_fog: Float32,
+    contrast_power: Float32,
+    skew_power: Float32,
+    hue_preservation: Float32,
+    pipe_to_base: SIMD[DType.float32, 16],
+    base_to_rendering: SIMD[DType.float32, 16],
+    rendering_to_pipe: SIMD[DType.float32, 16],
+) -> SIMD[DType.float32, 4]:
+    var i_r = pipe_to_base[0] * in_r + pipe_to_base[1] * in_g + pipe_to_base[2] * in_b
+    var i_g = pipe_to_base[4] * in_r + pipe_to_base[5] * in_g + pipe_to_base[6] * in_b
+    var i_b = pipe_to_base[8] * in_r + pipe_to_base[9] * in_g + pipe_to_base[10] * in_b
+
+    var avg = max((i_r + i_g + i_b) / Float32(3.0), Float32(0.0))
+    var min_v = min(min(i_r, i_g), i_b)
+    var sat = Float32(1.0)
+    if min_v < Float32(0.0):
+        sat = -avg / (min_v - avg)
+    i_r = avg + sat * (i_r - avg)
+    i_g = avg + sat * (i_g - avg)
+    i_b = avg + sat * (i_b - avg)
+
+    var r_r = base_to_rendering[0] * i_r + base_to_rendering[1] * i_g + base_to_rendering[2] * i_b
+    var r_g = base_to_rendering[4] * i_r + base_to_rendering[5] * i_g + base_to_rendering[6] * i_b
+    var r_b = base_to_rendering[8] * i_r + base_to_rendering[9] * i_g + base_to_rendering[10] * i_b
+
+    var pc_r = generalized_loglogistic_sigmoid_scalar(
+        r_r, white_target, paper_exp, film_fog, contrast_power, skew_power
+    )
+    var pc_g = generalized_loglogistic_sigmoid_scalar(
+        r_g, white_target, paper_exp, film_fog, contrast_power, skew_power
+    )
+    var pc_b = generalized_loglogistic_sigmoid_scalar(
+        r_b, white_target, paper_exp, film_fog, contrast_power, skew_power
+    )
+
+    var p_min: Float32
+    var p_mid: Float32
+    var p_max: Float32
+    var pc_min: Float32
+    var pc_mid: Float32
+    var pc_max: Float32
+
+    if r_r >= r_g:
+        if r_g >= r_b:
+            p_max = r_r; p_mid = r_g; p_min = r_b
+            pc_max = pc_r; pc_mid = pc_g; pc_min = pc_b
+        elif r_b >= r_r:
+            p_max = r_b; p_mid = r_r; p_min = r_g
+            pc_max = pc_b; pc_mid = pc_r; pc_min = pc_g
+        else:
+            p_max = r_r; p_mid = r_b; p_min = r_g
+            pc_max = pc_r; pc_mid = pc_b; pc_min = pc_g
+    else:
+        if r_r >= r_b:
+            p_max = r_g; p_mid = r_r; p_min = r_b
+            pc_max = pc_g; pc_mid = pc_r; pc_min = pc_b
+        elif r_b >= r_g:
+            p_max = r_b; p_mid = r_g; p_min = r_r
+            pc_max = pc_b; pc_mid = pc_g; pc_min = pc_r
+        else:
+            p_max = r_g; p_mid = r_b; p_min = r_r
+            pc_max = pc_g; pc_mid = pc_b; pc_min = pc_r
+
+    var chroma = p_max - p_min
+    var midscale = Float32(0.0)
+    if chroma != Float32(0.0):
+        midscale = (p_mid - p_min) / chroma
+
+    var f_hc = pc_min + (pc_max - pc_min) * midscale
+    var n_mid = (Float32(1.0) - hue_preservation) * pc_mid + hue_preservation * f_hc
+
+    var blend = Float32(2.0) * p_min / (p_min + p_mid + Float32(1e-9))
+    var target = blend * (pc_r + pc_g + pc_b) + (Float32(1.0) - blend) * (
+        pc_min + n_mid + pc_max
+    )
+
+    var res_min: Float32
+    var res_mid: Float32
+    var res_max: Float32
+    if n_mid <= pc_mid:
+        res_mid = (
+            (Float32(1.0) - hue_preservation) * pc_mid
+            + hue_preservation
+            * (midscale * pc_max + (Float32(1.0) - midscale) * (target - pc_max))
+        ) / (Float32(1.0) + hue_preservation * (Float32(1.0) - midscale))
+        res_min = target - pc_max - res_mid
+        res_max = pc_max
+    else:
+        res_mid = (
+            (Float32(1.0) - hue_preservation) * pc_mid
+            + hue_preservation
+            * (pc_min * (Float32(1.0) - midscale) + midscale * (target - pc_min))
+        ) / (Float32(1.0) + hue_preservation * midscale)
+        res_min = pc_min
+        res_max = target - pc_min - res_mid
+
+    var res_r: Float32
+    var res_g: Float32
+    var res_b: Float32
+    if r_r >= r_g:
+        if r_g >= r_b:
+            res_r = res_max; res_g = res_mid; res_b = res_min
+        elif r_b >= r_r:
+            res_b = res_max; res_r = res_mid; res_g = res_min
+        else:
+            res_r = res_max; res_b = res_mid; res_g = res_min
+    else:
+        if r_r >= r_b:
+            res_g = res_max; res_r = res_mid; res_b = res_min
+        elif r_b >= r_g:
+            res_b = res_max; res_g = res_mid; res_r = res_min
+        else:
+            res_g = res_max; res_b = res_mid; res_r = res_min
+
+    var out_r = rendering_to_pipe[0] * res_r + rendering_to_pipe[1] * res_g + rendering_to_pipe[2] * res_b
+    var out_g = rendering_to_pipe[4] * res_r + rendering_to_pipe[5] * res_g + rendering_to_pipe[6] * res_b
+    var out_b = rendering_to_pipe[8] * res_r + rendering_to_pipe[9] * res_g + rendering_to_pipe[10] * res_b
+
+    return SIMD[DType.float32, 4](out_r, out_g, out_b, in_a)
diff --git a/mojo/iop/sigmoid/lib.mojo b/mojo/iop/sigmoid/lib.mojo
new file mode 100644
index 0000000000..3b9191e0f6
--- /dev/null
+++ b/mojo/iop/sigmoid/lib.mojo
@@ -0,0 +1,432 @@
+from std.gpu.host import DeviceContext
+from layout import Layout, LayoutTensor, UNKNOWN_VALUE
+from layout.runtime_layout import RuntimeLayout, IndexList
+from layout.coord import Coord
+from std.utils import Index
+from std.algorithm.functional import elementwise
+from std.memory import alloc
+from std.memory.unsafe_pointer import UnsafePointer
+from iop.sigmoid.kernels import (
+    apply_sigmoid_rgb_ratio,
+    apply_sigmoid_per_channel,
+)
+
+comptime CHANNELS = 4
+comptime IMAGE_LAYOUT = Layout.row_major(UNKNOWN_VALUE, UNKNOWN_VALUE, CHANNELS)
+comptime DTYPE = DType.float32
+
+
+struct MojoCtx:
+    var use_gpu: Int
+    var dctx_addr: Int
+
+    def __init__(out self, use_gpu: Int, dctx_addr: Int):
+        self.use_gpu = use_gpu
+        self.dctx_addr = dctx_addr
+
+
+struct CParams:
+    var wt: Float32
+    var bt: Float32
+    var pe: Float32
+    var ff: Float32
+    var fp: Float32
+    var pp: Float32
+    var hp: Float32
+    var ptb: SIMD[DType.float32, 16]
+    var btr: SIMD[DType.float32, 16]
+    var rtp: SIMD[DType.float32, 16]
+
+    def __init__(out self, addr: Int):
+        var p = UnsafePointer[Float32, MutAnyOrigin](unsafe_from_address=addr)
+        self.wt = p[0]
+        self.bt = p[4]
+        self.pe = p[8]
+        self.ff = p[12]
+        self.fp = p[16]
+        self.pp = p[20]
+        self.hp = p[32]
+        var m_ptb = SIMD[DType.float32, 16]()
+        var m_btr = SIMD[DType.float32, 16]()
+        var m_rtp = SIMD[DType.float32, 16]()
+        for i in range(16):
+            m_ptb[i] = p[36 + i]
+            m_btr[i] = p[52 + i]
+            m_rtp[i] = p[68 + i]
+        self.ptb = m_ptb
+        self.btr = m_btr
+        self.rtp = m_rtp
+
+
+def _launch_rgb_ratio_gpu(
+    dctx: DeviceContext,
+    in_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin],
+    out_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
+    wt: Float32,
+    bt: Float32,
+    pe: Float32,
+    ff: Float32,
+    fp: Float32,
+    pp: Float32,
+    img_w: Int,
+    img_h: Int,
+    num_pixels: Int,
+) raises:
+    @parameter
+    @always_inline
+    def gpu_kernel[simd_width: Int, alignment: Int](coord: Coord) capturing -> None:
+        var px_idx = Int(coord[0].value())
+        var y = px_idx // img_w
+        var x = px_idx % img_w
+        var pix = in_t.load[width=4](Index(y, x, 0))
+        var r = apply_sigmoid_rgb_ratio(
+            pix[0],
+            pix[1],
+            pix[2],
+            pix[3],
+            wt,
+            bt,
+            pe,
+            ff,
+            fp,
+            pp,
+        )
+        out_t.store[width=4](Index(y, x, 0), r)
+
+    elementwise[gpu_kernel, 1, target="gpu"](num_pixels, dctx)
+    dctx.synchronize()
+
+
+def _launch_per_channel_gpu(
+    dctx: DeviceContext,
+    in_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin],
+    out_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
+    wt: Float32,
+    pe: Float32,
+    ff: Float32,
+    fp: Float32,
+    pp: Float32,
+    hp: Float32,
+    kptb: SIMD[DType.float32, 16],
+    kbtr: SIMD[DType.float32, 16],
+    krtp: SIMD[DType.float32, 16],
+    img_w: Int,
+    img_h: Int,
+    num_pixels: Int,
+) raises:
+    @parameter
+    @always_inline
+    def gpu_kernel[simd_width: Int, alignment: Int](coord: Coord) capturing -> None:
+        var px_idx = Int(coord[0].value())
+        var y = px_idx // img_w
+        var x = px_idx % img_w
+        var pix = in_t.load[width=4](Index(y, x, 0))
+        var r = apply_sigmoid_per_channel(
+            pix[0],
+            pix[1],
+            pix[2],
+            pix[3],
+            wt,
+            pe,
+            ff,
+            fp,
+            pp,
+            hp,
+            kptb,
+            kbtr,
+            krtp,
+        )
+        out_t.store[width=4](Index(y, x, 0), r)
+
+    elementwise[gpu_kernel, 1, target="gpu"](num_pixels, dctx)
+    dctx.synchronize()
+
+
+@export("sigmoid_mojo_init")
+def sigmoid_mojo_init(ctx_out: UnsafePointer[Int, MutAnyOrigin], use_gpu: Int32) abi("C") -> None:
+    var gpu = use_gpu != 0
+    var dctx_addr = 0
+    if gpu:
+        try:
+            var d_ptr = alloc[DeviceContext](1)
+            d_ptr.unsafe_write(DeviceContext())
+            dctx_addr = Int(d_ptr)
+            print("Mojo: GPU Context Initialized Successfully")
+        except e:
+            print("Mojo: GPU Init Error (falling back to CPU):", String(e))
+            gpu = False
+    else:
+        print("Mojo: CPU Context Initialized")
+    var p = alloc[MojoCtx](1)
+    p[0].use_gpu = 1 if gpu else 0
+    p[0].dctx_addr = dctx_addr
+    ctx_out[0] = Int(p)
+
+
+@export("sigmoid_mojo_destroy")
+def sigmoid_mojo_destroy(ctx_addr: Int) abi("C") -> None:
+    var p = UnsafePointer[MojoCtx, MutAnyOrigin](unsafe_from_address=ctx_addr)
+    if p[0].dctx_addr != 0:
+        var dctx_ptr = UnsafePointer[DeviceContext, MutAnyOrigin](
+            unsafe_from_address=p[0].dctx_addr
+        )
+        dctx_ptr.unsafe_deinit_pointee()
+        dctx_ptr.free()
+    p.free()
+
+
+def _run_cpu_rgb_ratio(
+    in_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
+    out_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
+    wt: Float32,
+    bt: Float32,
+    pe: Float32,
+    ff: Float32,
+    fp: Float32,
+    pp: Float32,
+    width: Int,
+    height: Int,
+    num_pixels: Int,
+):
+    for i in range(num_pixels):
+        var px = in_t.load[width=4](Index(i // width, i % width, 0))
+        var r = apply_sigmoid_rgb_ratio(
+            px[0], px[1], px[2], px[3], wt, bt, pe, ff, fp, pp
+        )
+        out_t.store[width=4](Index(i // width, i % width, 0), r)
+
+
+def _run_cpu_per_channel(
+    in_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
+    out_t: LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin],
+    wt: Float32,
+    pe: Float32,
+    ff: Float32,
+    fp: Float32,
+    pp: Float32,
+    hp: Float32,
+    ptb: SIMD[DType.float32, 16],
+    btr: SIMD[DType.float32, 16],
+    rtp: SIMD[DType.float32, 16],
+    width: Int,
+    height: Int,
+    num_pixels: Int,
+):
+    for i in range(num_pixels):
+        var px = in_t.load[width=4](Index(i // width, i % width, 0))
+        var r = apply_sigmoid_per_channel(
+            px[0],
+            px[1],
+            px[2],
+            px[3],
+            wt,
+            pe,
+            ff,
+            fp,
+            pp,
+            hp,
+            ptb,
+            btr,
+            rtp,
+        )
+        out_t.store[width=4](Index(i // width, i % width, 0), r)
+
+
+@export("sigmoid_mojo_rgb_ratio")
+def sigmoid_mojo_rgb_ratio(
+    ctx_addr: Int,
+    in_addr: Int,
+    out_addr: Int,
+    width: Int32,
+    height: Int32,
+    p_addr: Int,
+) abi("C") -> None:
+    var ctx_p = UnsafePointer[MojoCtx, MutAnyOrigin](
+        unsafe_from_address=ctx_addr
+    )
+    var use_gpu = ctx_p[0].use_gpu != 0
+    var dctx_addr = ctx_p[0].dctx_addr
+    var params = CParams(p_addr)
+    var h = Int(height)
+    var w = Int(width)
+    var num_pixels = h * w
+
+    if use_gpu and dctx_addr != 0:
+        try:
+            var dctx = UnsafePointer[DeviceContext, MutAnyOrigin](
+                unsafe_from_address=dctx_addr
+            )[0]
+            var dev_in = dctx.enqueue_create_buffer[DTYPE](num_pixels * 4)
+            var dev_out = dctx.enqueue_create_buffer[DTYPE](num_pixels * 4)
+            var in_p = UnsafePointer[Float32, MutAnyOrigin](
+                unsafe_from_address=in_addr
+            )
+            dctx.enqueue_copy(dev_in, in_p)
+
+            var rt_gpu = RuntimeLayout[IMAGE_LAYOUT].row_major(
+                IndexList[3](h, w, CHANNELS)
+            )
+            var in_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin](
+                UnsafePointer[Float32, ImmutAnyOrigin](
+                    unsafe_from_address=Int(dev_in.unsafe_ptr())
+                ),
+                rt_gpu,
+            )
+            var out_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
+                UnsafePointer[Float32, MutAnyOrigin](
+                    unsafe_from_address=Int(dev_out.unsafe_ptr())
+                ),
+                rt_gpu,
+            )
+
+            _launch_rgb_ratio_gpu(
+                dctx,
+                in_t,
+                out_t,
+                params.wt,
+                params.bt,
+                params.pe,
+                params.ff,
+                params.fp,
+                params.pp,
+                w,
+                h,
+                num_pixels,
+            )
+
+            var out_p = UnsafePointer[Float32, MutAnyOrigin](
+                unsafe_from_address=out_addr
+            )
+            dctx.enqueue_copy(out_p, dev_out)
+            dctx.synchronize()
+        except e:
+            print("GPU Run Error (RGB Ratio):", String(e))
+    else:
+        var rt = RuntimeLayout[IMAGE_LAYOUT].row_major(
+            IndexList[3](h, w, CHANNELS)
+        )
+        var in_p = UnsafePointer[Float32, MutAnyOrigin](
+            unsafe_from_address=in_addr
+        )
+        var out_p = UnsafePointer[Float32, MutAnyOrigin](
+            unsafe_from_address=out_addr
+        )
+        var in_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](in_p, rt)
+        var out_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](out_p, rt)
+
+        _run_cpu_rgb_ratio(
+            in_t,
+            out_t,
+            params.wt,
+            params.bt,
+            params.pe,
+            params.ff,
+            params.fp,
+            params.pp,
+            w,
+            h,
+            num_pixels,
+        )
+
+
+@export("sigmoid_mojo_per_channel")
+def sigmoid_mojo_per_channel(
+    ctx_addr: Int,
+    in_addr: Int,
+    out_addr: Int,
+    width: Int32,
+    height: Int32,
+    p_addr: Int,
+) abi("C") -> None:
+    var ctx_p = UnsafePointer[MojoCtx, MutAnyOrigin](
+        unsafe_from_address=ctx_addr
+    )
+    var use_gpu = ctx_p[0].use_gpu != 0
+    var dctx_addr = ctx_p[0].dctx_addr
+    var params = CParams(p_addr)
+    var h = Int(height)
+    var w = Int(width)
+    var num_pixels = h * w
+
+    if use_gpu and dctx_addr != 0:
+        try:
+            var dctx = UnsafePointer[DeviceContext, MutAnyOrigin](
+                unsafe_from_address=dctx_addr
+            )[0]
+            var dev_in = dctx.enqueue_create_buffer[DTYPE](num_pixels * 4)
+            var dev_out = dctx.enqueue_create_buffer[DTYPE](num_pixels * 4)
+            var in_p = UnsafePointer[Float32, MutAnyOrigin](
+                unsafe_from_address=in_addr
+            )
+            dctx.enqueue_copy(dev_in, in_p)
+
+            var rt_gpu = RuntimeLayout[IMAGE_LAYOUT].row_major(
+                IndexList[3](h, w, CHANNELS)
+            )
+            var in_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, ImmutAnyOrigin](
+                UnsafePointer[Float32, ImmutAnyOrigin](
+                    unsafe_from_address=Int(dev_in.unsafe_ptr())
+                ),
+                rt_gpu,
+            )
+            var out_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](
+                UnsafePointer[Float32, MutAnyOrigin](
+                    unsafe_from_address=Int(dev_out.unsafe_ptr())
+                ),
+                rt_gpu,
+            )
+
+            _launch_per_channel_gpu(
+                dctx,
+                in_t,
+                out_t,
+                params.wt,
+                params.pe,
+                params.ff,
+                params.fp,
+                params.pp,
+                params.hp,
+                params.ptb,
+                params.btr,
+                params.rtp,
+                w,
+                h,
+                num_pixels,
+            )
+
+            var out_p = UnsafePointer[Float32, MutAnyOrigin](
+                unsafe_from_address=out_addr
+            )
+            dctx.enqueue_copy(out_p, dev_out)
+            dctx.synchronize()
+        except e:
+            print("GPU Error (Per Channel):", String(e))
+    else:
+        var rt = RuntimeLayout[IMAGE_LAYOUT].row_major(
+            IndexList[3](h, w, CHANNELS)
+        )
+        var in_p = UnsafePointer[Float32, MutAnyOrigin](
+            unsafe_from_address=in_addr
+        )
+        var out_p = UnsafePointer[Float32, MutAnyOrigin](
+            unsafe_from_address=out_addr
+        )
+        var in_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](in_p, rt)
+        var out_t = LayoutTensor[DTYPE, IMAGE_LAYOUT, MutAnyOrigin](out_p, rt)
+
+        _run_cpu_per_channel(
+            in_t,
+            out_t,
+            params.wt,
+            params.pe,
+            params.ff,
+            params.fp,
+            params.pp,
+            params.hp,
+            params.ptb,
+            params.btr,
+            params.rtp,
+            w,
+            h,
+            num_pixels,
+        )
diff --git a/benchmark/pixi.lock b/mojo/pixi.lock
similarity index 51%
rename from benchmark/pixi.lock
rename to mojo/pixi.lock
index 38ce7ccdb5..f4dd8df35b 100644
--- a/benchmark/pixi.lock
+++ b/mojo/pixi.lock
@@ -1,82 +1,87 @@
-version: 6
+version: 7
+platforms:
+- name: linux-64
+  virtual-packages:
+  - __unix=0=0
+  - __linux=4.18
+  - __glibc=2.28
+  - __archspec=0=x86_64
 environments:
   default:
     channels:
     - url: https://conda.modular.com/max-nightly/
     - url: https://conda.anaconda.org/conda-forge/
-    options:
-      pypi-prerelease-mode: if-necessary-or-explicit
     packages:
       linux-64:
       - conda: https://conda.anaconda.org/conda-forge/linux-64/_openmp_mutex-4.5-20_gnu.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/_python_abi3_support-1.0-hd8ed1ab_2.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/bzip2-1.0.8-hda65f42_8.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/ca-certificates-2026.1.4-hbd8a1cb_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/click-8.3.1-pyh8f84b5b_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/cpython-3.14.3-py314hd8ed1ab_101.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/icu-78.2-h33c6efd_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/importlib-metadata-8.7.0-pyhe01879c_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_client-8.6.3-pyhd8ed1ab_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_core-5.9.1-pyhc90fa1f_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/bzip2-1.0.8-hda65f42_9.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/icu-78.3-h54a6638_2.conda
       - conda: https://conda.anaconda.org/conda-forge/linux-64/keyutils-1.6.3-hb9d3cd8_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/krb5-1.21.3-h659f571_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/ld_impl_linux-64-2.45.1-default_hbd61a6d_101.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libblas-3.11.0-5_h4a7cf45_openblas.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libcblas-3.11.0-5_h0358290_openblas.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/krb5-1.22.2-hbde042b_1.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/ld_impl_linux-64-2.46.1-default_hbd61a6d_102.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libblas-3.11.0-8_h4a7cf45_openblas.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libcblas-3.11.0-8_h0358290_openblas.conda
       - conda: https://conda.anaconda.org/conda-forge/linux-64/libedit-3.1.20250104-pl5321h7949ede_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libexpat-2.7.3-hecca717_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libexpat-2.8.1-hecca717_1.conda
       - conda: https://conda.anaconda.org/conda-forge/linux-64/libffi-3.5.2-h3435931_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgcc-15.2.0-he0feb66_17.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgcc-ng-15.2.0-h69a702a_17.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran-15.2.0-h69a702a_17.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran5-15.2.0-h68bc16d_17.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgomp-15.2.0-he0feb66_17.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/liblapack-3.11.0-5_h47877c9_openblas.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/liblzma-5.8.2-hb03c661_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgcc-15.2.0-he0feb66_20.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran-15.2.0-h69a702a_20.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran5-15.2.0-h68bc16d_20.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libgomp-15.2.0-he0feb66_20.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/liblapack-3.11.0-8_h47877c9_openblas.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/liblzma-5.8.3-hb03c661_0.conda
       - conda: https://conda.anaconda.org/conda-forge/linux-64/libmpdec-4.0.0-hb03c661_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libopenblas-0.3.30-pthreads_h94d23a6_4.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libsodium-1.0.20-h4ab18f5_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libsqlite-3.51.2-hf4e2dac_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libstdcxx-15.2.0-h934c35e_17.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libstdcxx-ng-15.2.0-hdf11a46_17.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libuuid-2.41.3-h5347b49_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/libzlib-1.3.1-hb9d3cd8_2.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/markdown-it-py-4.0.0-pyhd8ed1ab_0.conda
-      - conda: https://conda.modular.com/max-nightly/linux-64/max-26.2.0.dev2026022205-3.14release.conda
-      - conda: https://conda.modular.com/max-nightly/linux-64/max-core-26.2.0.dev2026022205-release.conda
-      - conda: https://conda.modular.com/max-nightly/noarch/mblack-26.2.0.dev2026022205-release.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libopenblas-0.3.33-pthreads_h94d23a6_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libsodium-1.0.22-h280c20c_1.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libsqlite-3.53.4-hf4e2dac_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libstdcxx-15.2.0-h934c35e_20.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libuuid-2.42.2-h5347b49_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/libzlib-1.3.2-h25fd6f3_2.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/msgspec-0.21.1-py314h5bd0f2a_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/ncurses-6.6-hdb14827_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/numpy-2.5.1-py314h2b28147_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/openssl-3.6.3-h35e630c_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/psutil-7.2.2-py314h0f05182_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/python-3.14.6-habeac84_101_cp314.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/pyzmq-27.1.0-py312hda471dd_3.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/readline-8.3-h853b02a_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/tk-8.6.13-noxft_hd70dff1_3.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/tornado-6.5.7-py314h5bd0f2a_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/zeromq-4.3.5-h09e67af_11.conda
+      - conda: https://conda.anaconda.org/conda-forge/linux-64/zstd-1.5.7-hb78ec9c_6.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/_python_abi3_support-1.0-hd8ed1ab_3.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/ca-certificates-2026.7.22-hbd8a1cb_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/click-8.4.2-pyhc90fa1f_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/cpython-3.14.6-py314hd8ed1ab_101.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/exceptiongroup-1.3.1-pyhd8ed1ab_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/importlib-metadata-9.0.0-pyhcf101f3_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_client-8.6.3-pyhd8ed1ab_1.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_core-5.9.1-pyhc90fa1f_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/markdown-it-py-4.2.0-pyhd8ed1ab_0.conda
       - conda: https://conda.anaconda.org/conda-forge/noarch/mdurl-0.1.2-pyhd8ed1ab_1.conda
-      - conda: https://conda.modular.com/max-nightly/linux-64/mojo-0.26.2.0.dev2026022205-release.conda
-      - conda: https://conda.modular.com/max-nightly/linux-64/mojo-compiler-0.26.2.0.dev2026022205-release.conda
-      - conda: https://conda.modular.com/max-nightly/noarch/mojo-python-0.26.2.0.dev2026022205-release.conda
       - conda: https://conda.anaconda.org/conda-forge/noarch/mypy_extensions-1.1.0-pyha770c72_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/ncurses-6.5-h2d0b736_3.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/numpy-2.4.2-py314h2b28147_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/openssl-3.6.1-h35e630c_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/packaging-26.0-pyhcf101f3_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/pathspec-1.0.4-pyhd8ed1ab_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/platformdirs-4.7.0-pyhcf101f3_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/pygments-2.19.2-pyhd8ed1ab_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/python-3.14.3-h32b2ec7_101_cp314.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/packaging-26.2-pyhc364b38_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/pathspec-1.1.1-pyhd8ed1ab_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/platformdirs-4.11.0-pyhcf101f3_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/pygments-2.20.0-pyhd8ed1ab_0.conda
       - conda: https://conda.anaconda.org/conda-forge/noarch/python-dateutil-2.9.0.post0-pyhe01879c_2.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/python-gil-3.14.3-h4df99d1_101.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/python-gil-3.14.6-h4df99d1_101.conda
       - conda: https://conda.anaconda.org/conda-forge/noarch/python_abi-3.14-8_cp314.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/pyyaml-6.0.3-py314h67df5f8_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/pyzmq-27.1.0-py312hfb55c3c_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/readline-8.3-h853b02a_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/rich-14.3.2-pyhcf101f3_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/rich-15.0.0-pyhcf101f3_0.conda
       - conda: https://conda.anaconda.org/conda-forge/noarch/six-1.17.0-pyhe01879c_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/tk-8.6.13-noxft_h366c992_103.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/tomli-2.4.0-pyhcf101f3_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/tornado-6.5.3-py314h5bd0f2a_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/traitlets-5.14.3-pyhd8ed1ab_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/typing-extensions-4.15.0-h396c80c_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/typing_extensions-4.15.0-pyhcf101f3_0.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/tzdata-2025c-hc9c84f9_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/yaml-0.2.5-h280c20c_3.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/zeromq-4.3.5-h387f397_9.conda
-      - conda: https://conda.anaconda.org/conda-forge/noarch/zipp-3.23.0-pyhcf101f3_1.conda
-      - conda: https://conda.anaconda.org/conda-forge/linux-64/zstd-1.5.7-hb78ec9c_6.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/taskgroup-0.2.2-pyhd8ed1ab_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/tomli-2.4.1-pyhcf101f3_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/traitlets-5.15.1-pyhcf101f3_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/typing-extensions-4.16.0-h69aa097_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/typing_extensions-4.16.0-pyhcf101f3_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/tzdata-2026c-h151e31d_0.conda
+      - conda: https://conda.anaconda.org/conda-forge/noarch/zipp-4.1.0-pyhcf101f3_0.conda
+      - conda: https://conda.modular.com/max-nightly/linux-64/max-26.5.0.dev2026072505-3.14release.conda
+      - conda: https://conda.modular.com/max-nightly/linux-64/max-core-26.5.0.dev2026072505-release.conda
+      - conda: https://conda.modular.com/max-nightly/linux-64/mojo-1.0.0b3.dev2026072505-release.conda
+      - conda: https://conda.modular.com/max-nightly/linux-64/mojo-compiler-1.0.0b3.dev2026072505-release.conda
+      - conda: https://conda.modular.com/max-nightly/noarch/mblack-26.5.0.dev2026072505-release.conda
+      - conda: https://conda.modular.com/max-nightly/noarch/mojo-python-1.0.0b3.dev2026072505-release.conda
 packages:
 - conda: https://conda.anaconda.org/conda-forge/linux-64/_openmp_mutex-4.5-20_gnu.conda
   build_number: 20
@@ -88,110 +93,39 @@ packages:
   constrains:
   - openmp_impl <0.0a0
   license: BSD-3-Clause
+  license_family: BSD
+  run_exports:
+    strong:
+    - _openmp_mutex >=4.5
   size: 28948
   timestamp: 1770939786096
-- conda: https://conda.anaconda.org/conda-forge/noarch/_python_abi3_support-1.0-hd8ed1ab_2.conda
-  sha256: a3967b937b9abf0f2a99f3173fa4630293979bd1644709d89580e7c62a544661
-  md5: aaa2a381ccc56eac91d63b6c1240312f
-  depends:
-  - cpython
-  - python-gil
-  license: MIT
-  license_family: MIT
-  size: 8191
-  timestamp: 1744137672556
-- conda: https://conda.anaconda.org/conda-forge/linux-64/bzip2-1.0.8-hda65f42_8.conda
-  sha256: c30daba32ddebbb7ded490f0e371eae90f51e72db620554089103b4a6934b0d5
-  md5: 51a19bba1b8ebfb60df25cde030b7ebc
+- conda: https://conda.anaconda.org/conda-forge/linux-64/bzip2-1.0.8-hda65f42_9.conda
+  sha256: 0b75d45f0bba3e95dc693336fa51f40ea28c980131fec438afb7ce6118ed05f6
+  md5: d2ffd7602c02f2b316fd921d39876885
   depends:
   - __glibc >=2.17,<3.0.a0
   - libgcc >=14
   license: bzip2-1.0.6
   license_family: BSD
-  size: 260341
-  timestamp: 1757437258798
-- conda: https://conda.anaconda.org/conda-forge/noarch/ca-certificates-2026.1.4-hbd8a1cb_0.conda
-  sha256: b5974ec9b50e3c514a382335efa81ed02b05906849827a34061c496f4defa0b2
-  md5: bddacf101bb4dd0e51811cb69c7790e2
+  run_exports:
+    weak:
+    - bzip2 >=1.0.8,<2.0a0
+  size: 260182
+  timestamp: 1771350215188
+- conda: https://conda.anaconda.org/conda-forge/linux-64/icu-78.3-h54a6638_2.conda
+  sha256: d7c260b7e1cf22ce04d6ba8a86eabf4e6c50bc96a5c27fe2ecb32298af3e88eb
+  md5: 4ef4b977bb216a3001a3334696a80850
   depends:
-  - __unix
-  license: ISC
-  size: 146519
-  timestamp: 1767500828366
-- conda: https://conda.anaconda.org/conda-forge/noarch/click-8.3.1-pyh8f84b5b_1.conda
-  sha256: 38cfe1ee75b21a8361c8824f5544c3866f303af1762693a178266d7f198e8715
-  md5: ea8a6c3256897cc31263de9f455e25d9
-  depends:
-  - python >=3.10
-  - __unix
-  - python
-  license: BSD-3-Clause
-  license_family: BSD
-  size: 97676
-  timestamp: 1764518652276
-- conda: https://conda.anaconda.org/conda-forge/noarch/cpython-3.14.3-py314hd8ed1ab_101.conda
-  noarch: generic
-  sha256: 91b06300879df746214f7363d6c27c2489c80732e46a369eb2afc234bcafb44c
-  md5: 3bb89e4f795e5414addaa531d6b1500a
-  depends:
-  - python >=3.14,<3.15.0a0
-  - python_abi * *_cp314
-  license: Python-2.0
-  size: 50078
-  timestamp: 1770674447292
-- conda: https://conda.anaconda.org/conda-forge/linux-64/icu-78.2-h33c6efd_0.conda
-  sha256: 142a722072fa96cf16ff98eaaf641f54ab84744af81754c292cb81e0881c0329
-  md5: 186a18e3ba246eccfc7cff00cd19a870
-  depends:
-  - __glibc >=2.17,<3.0.a0
-  - libgcc >=14
   - libstdcxx >=14
+  - libgcc >=14
+  - __glibc >=2.17,<3.0.a0
   license: MIT
   license_family: MIT
-  size: 12728445
-  timestamp: 1767969922681
-- conda: https://conda.anaconda.org/conda-forge/noarch/importlib-metadata-8.7.0-pyhe01879c_1.conda
-  sha256: c18ab120a0613ada4391b15981d86ff777b5690ca461ea7e9e49531e8f374745
-  md5: 63ccfdc3a3ce25b027b8767eb722fca8
-  depends:
-  - python >=3.9
-  - zipp >=3.20
-  - python
-  license: Apache-2.0
-  license_family: APACHE
-  size: 34641
-  timestamp: 1747934053147
-- conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_client-8.6.3-pyhd8ed1ab_1.conda
-  sha256: 19d8bd5bb2fde910ec59e081eeb59529491995ce0d653a5209366611023a0b3a
-  md5: 4ebae00eae9705b0c3d6d1018a81d047
-  depends:
-  - importlib-metadata >=4.8.3
-  - jupyter_core >=4.12,!=5.0.*
-  - python >=3.9
-  - python-dateutil >=2.8.2
-  - pyzmq >=23.0
-  - tornado >=6.2
-  - traitlets >=5.3
-  license: BSD-3-Clause
-  license_family: BSD
-  size: 106342
-  timestamp: 1733441040958
-- conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_core-5.9.1-pyhc90fa1f_0.conda
-  sha256: 1d34b80e5bfcd5323f104dbf99a2aafc0e5d823019d626d0dce5d3d356a2a52a
-  md5: b38fe4e78ee75def7e599843ef4c1ab0
-  depends:
-  - __unix
-  - python
-  - platformdirs >=2.5
-  - python >=3.10
-  - traitlets >=5.3
-  - python
-  constrains:
-  - pywin32 >=300
-  license: BSD-3-Clause
-  license_family: BSD
-  size: 65503
-  timestamp: 1760643864586
+  run_exports:
+    weak:
+    - icu >=78.3,<79.0a0
+  size: 14455340
+  timestamp: 1784916378180
 - conda: https://conda.anaconda.org/conda-forge/linux-64/keyutils-1.6.3-hb9d3cd8_0.conda
   sha256: 0960d06048a7185d3542d850986d807c6e37ca2e644342dd0c72feefcf26c2a4
   md5: b38117a3c920364aff79f870c984b4a3
@@ -199,65 +133,79 @@ packages:
   - __glibc >=2.17,<3.0.a0
   - libgcc >=13
   license: LGPL-2.1-or-later
+  run_exports:
+    weak:
+    - keyutils >=1.6.3,<2.0a0
   size: 134088
   timestamp: 1754905959823
-- conda: https://conda.anaconda.org/conda-forge/linux-64/krb5-1.21.3-h659f571_0.conda
-  sha256: 99df692f7a8a5c27cd14b5fb1374ee55e756631b9c3d659ed3ee60830249b238
-  md5: 3f43953b7d3fb3aaa1d0d0723d91e368
-  depends:
-  - keyutils >=1.6.1,<2.0a0
-  - libedit >=3.1.20191231,<3.2.0a0
-  - libedit >=3.1.20191231,<4.0a0
-  - libgcc-ng >=12
-  - libstdcxx-ng >=12
-  - openssl >=3.3.1,<4.0a0
+- conda: https://conda.anaconda.org/conda-forge/linux-64/krb5-1.22.2-hbde042b_1.conda
+  sha256: 9b07046870772f28740e3f6149f09ff222843733087a33c5540b169c6289652d
+  md5: 54157a1c8c0bb70f62dd0b17fba7e7f2
+  depends:
+  - __glibc >=2.17,<3.0.a0
+  - keyutils >=1.6.3,<2.0a0
+  - libedit >=3.1.20250104,<3.2.0a0
+  - libedit >=3.1.20250104,<4.0a0
+  - libgcc >=14
+  - libstdcxx >=14
+  - openssl >=3.5.7,<4.0a0
   license: MIT
   license_family: MIT
-  size: 1370023
-  timestamp: 1719463201255
-- conda: https://conda.anaconda.org/conda-forge/linux-64/ld_impl_linux-64-2.45.1-default_hbd61a6d_101.conda
-  sha256: 565941ac1f8b0d2f2e8f02827cbca648f4d18cd461afc31f15604cd291b5c5f3
-  md5: 12bd9a3f089ee6c9266a37dab82afabd
+  run_exports:
+    weak:
+    - krb5 >=1.22.2,<1.23.0a0
+  size: 1388990
+  timestamp: 1781859420533
+- conda: https://conda.anaconda.org/conda-forge/linux-64/ld_impl_linux-64-2.46.1-default_hbd61a6d_102.conda
+  sha256: 27d83f1188cd19bcb7754a078b3fa7f4cfb8527f8eb2fde54dd01fc529d1adec
+  md5: 449500f2c089da11c40f5c21312e3e07
   depends:
   - __glibc >=2.17,<3.0.a0
   - zstd >=1.5.7,<1.6.0a0
   constrains:
-  - binutils_impl_linux-64 2.45.1
+  - binutils_impl_linux-64 2.46.1
   license: GPL-3.0-only
   license_family: GPL
-  size: 725507
-  timestamp: 1770267139900
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libblas-3.11.0-5_h4a7cf45_openblas.conda
-  build_number: 5
-  sha256: 18c72545080b86739352482ba14ba2c4815e19e26a7417ca21a95b76ec8da24c
-  md5: c160954f7418d7b6e87eaf05a8913fa9
-  depends:
-  - libopenblas >=0.3.30,<0.3.31.0a0
-  - libopenblas >=0.3.30,<1.0a0
+  run_exports: {}
+  size: 745303
+  timestamp: 1784214507189
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libblas-3.11.0-8_h4a7cf45_openblas.conda
+  build_number: 8
+  sha256: b2da6bfd72a1c9cb143ccf64bf5b28790cb4eb58bd1cb978f6537b2322f7d48b
+  md5: 00fc660ab1b2f5ca07e92b4900d10c79
+  depends:
+  - libopenblas >=0.3.33,<0.3.34.0a0
+  - libopenblas >=0.3.33,<1.0a0
   constrains:
-  - mkl <2026
-  - liblapack  3.11.0   5*_openblas
-  - libcblas   3.11.0   5*_openblas
-  - blas 2.305   openblas
-  - liblapacke 3.11.0   5*_openblas
+  - blas 2.308   openblas
+  - mkl <2027
+  - libcblas   3.11.0   8*_openblas
+  - liblapack  3.11.0   8*_openblas
+  - liblapacke 3.11.0   8*_openblas
   license: BSD-3-Clause
   license_family: BSD
-  size: 18213
-  timestamp: 1765818813880
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libcblas-3.11.0-5_h0358290_openblas.conda
-  build_number: 5
-  sha256: 0cbdcc67901e02dc17f1d19e1f9170610bd828100dc207de4d5b6b8ad1ae7ad8
-  md5: 6636a2b6f1a87572df2970d3ebc87cc0
-  depends:
-  - libblas 3.11.0 5_h4a7cf45_openblas
+  run_exports:
+    weak:
+    - libblas >=3.11.0,<4.0a0
+  size: 18804
+  timestamp: 1779859100675
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libcblas-3.11.0-8_h0358290_openblas.conda
+  build_number: 8
+  sha256: 1a2bc77bb26520255904a3d9b1f40e6bf0bf9d8d3405c7709dd162282820915a
+  md5: 33a413f1095f8325e5c30fde3b0d2445
+  depends:
+  - libblas 3.11.0 8_h4a7cf45_openblas
   constrains:
-  - liblapacke 3.11.0   5*_openblas
-  - blas 2.305   openblas
-  - liblapack  3.11.0   5*_openblas
+  - blas 2.308   openblas
+  - liblapacke 3.11.0   8*_openblas
+  - liblapack  3.11.0   8*_openblas
   license: BSD-3-Clause
   license_family: BSD
-  size: 18194
-  timestamp: 1765818837135
+  run_exports:
+    weak:
+    - libcblas >=3.11.0,<4.0a0
+  size: 18778
+  timestamp: 1779859107964
 - conda: https://conda.anaconda.org/conda-forge/linux-64/libedit-3.1.20250104-pl5321h7949ede_0.conda
   sha256: d789471216e7aba3c184cd054ed61ce3f6dac6f87a50ec69291b9297f8c18724
   md5: c277e0a4d549b03ac1e9d6cbbe3d017b
@@ -268,20 +216,24 @@ packages:
   - ncurses >=6.5,<7.0a0
   license: BSD-2-Clause
   license_family: BSD
+  run_exports:
+    weak:
+    - libedit >=3.1.20250104,<3.2.0a0
   size: 134676
   timestamp: 1738479519902
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libexpat-2.7.3-hecca717_0.conda
-  sha256: 1e1b08f6211629cbc2efe7a5bca5953f8f6b3cae0eeb04ca4dacee1bd4e2db2f
-  md5: 8b09ae86839581147ef2e5c5e229d164
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libexpat-2.8.1-hecca717_1.conda
+  sha256: 16feffd9ddbbe5b718515d38ee376c685ba95491cd901244e24671d20b952a77
+  md5: b24d3c612f71e7aa74158d92106318b2
   depends:
   - __glibc >=2.17,<3.0.a0
   - libgcc >=14
   constrains:
-  - expat 2.7.3.*
+  - expat 2.8.1.*
   license: MIT
   license_family: MIT
-  size: 76643
-  timestamp: 1763549731408
+  run_exports: {}
+  size: 77856
+  timestamp: 1781203599810
 - conda: https://conda.anaconda.org/conda-forge/linux-64/libffi-3.5.2-h3435931_0.conda
   sha256: 31f19b6a88ce40ebc0d5a992c131f57d919f73c0b92cd1617a5bec83f6e961e6
   md5: a360c33a5abe61c07959e449fa1453eb
@@ -290,44 +242,40 @@ packages:
   - libgcc >=14
   license: MIT
   license_family: MIT
+  run_exports:
+    weak:
+    - libffi >=3.5.2,<3.6.0a0
   size: 58592
   timestamp: 1769456073053
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libgcc-15.2.0-he0feb66_17.conda
-  sha256: 43860222cf3abf04ded0cf24541a105aa388e0e1d4d6ca46258e186d4e87ae3e
-  md5: 3c281169ea25b987311400d7a7e28445
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libgcc-15.2.0-he0feb66_20.conda
+  sha256: e3b7bb287d1685b15781a6d9e3408d6ddaff23f5a1d0d6c0140619dd3950fe72
+  md5: 3533de187cf7283f96bfdb28ad73e2bc
   depends:
   - __glibc >=2.17,<3.0.a0
   - _openmp_mutex >=4.5
   constrains:
-  - libgcc-ng ==15.2.0=*_17
-  - libgomp 15.2.0 he0feb66_17
-  license: GPL-3.0-only WITH GCC-exception-3.1
-  license_family: GPL
-  size: 1040478
-  timestamp: 1770252533873
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libgcc-ng-15.2.0-h69a702a_17.conda
-  sha256: bdfe50501e4a2d904a5eae65a7ae26e2b7a29b473ab084ad55d96080b966502e
-  md5: 1478bfa85224a65ab096d69ffd2af1e5
-  depends:
-  - libgcc 15.2.0 he0feb66_17
+  - libgomp 15.2.0 he0feb66_20
+  - libgcc-ng ==15.2.0=*_20
   license: GPL-3.0-only WITH GCC-exception-3.1
   license_family: GPL
-  size: 27541
-  timestamp: 1770252546553
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran-15.2.0-h69a702a_17.conda
-  sha256: 1604c083dd65bc91e68b6cfe32c8610395088cb96af1acaf71f0dcaf83ac58f7
-  md5: a6c682ac611cb1fa4d73478f9e6efb06
-  depends:
-  - libgfortran5 15.2.0 h68bc16d_17
+  run_exports: {}
+  size: 1041122
+  timestamp: 1784923795784
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran-15.2.0-h69a702a_20.conda
+  sha256: 76d81032e264b74f519cc26962d5eb39547e0785fc9d2957bbc12e7a91214412
+  md5: a450a08a63f940e9aa7b37692e71196a
+  depends:
+  - libgfortran5 15.2.0 h68bc16d_20
   constrains:
-  - libgfortran-ng ==15.2.0=*_17
+  - libgfortran-ng ==15.2.0=*_20
   license: GPL-3.0-only WITH GCC-exception-3.1
   license_family: GPL
-  size: 27515
-  timestamp: 1770252591906
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran5-15.2.0-h68bc16d_17.conda
-  sha256: b1c77b85da9a3e204de986f59e262268805c6a35dffdf3953f1b98407db2aef3
-  md5: 202fdf8cad9eea704c2b0d823d1732bf
+  run_exports: {}
+  size: 28103
+  timestamp: 1784923831860
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libgfortran5-15.2.0-h68bc16d_20.conda
+  sha256: 95122f42566dc9a62b9615d2372b3f3c75fa7bd8bb1eda53aa7532ce60d545eb
+  md5: 4edbcbea1a8790a7d58e648523b69546
   depends:
   - __glibc >=2.17,<3.0.a0
   - libgcc >=15.2.0
@@ -335,42 +283,52 @@ packages:
   - libgfortran 15.2.0
   license: GPL-3.0-only WITH GCC-exception-3.1
   license_family: GPL
-  size: 2480824
-  timestamp: 1770252563579
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libgomp-15.2.0-he0feb66_17.conda
-  sha256: b961b5dd9761907a7179678b58a69bb4fc16b940eb477f635aea3aec0a3f17a6
-  md5: 51b78c6a757575c0d12f4401ffc67029
+  run_exports: {}
+  size: 2483727
+  timestamp: 1784923810904
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libgomp-15.2.0-he0feb66_20.conda
+  sha256: 30b8ff1bf43871a17c9de99e56171ef54cfbe690ffa86681628db5a00af97cf7
+  md5: 49321086c41bb58fc4b6cd8cbb74679d
   depends:
   - __glibc >=2.17,<3.0.a0
   license: GPL-3.0-only WITH GCC-exception-3.1
   license_family: GPL
-  size: 603334
-  timestamp: 1770252441199
-- conda: https://conda.anaconda.org/conda-forge/linux-64/liblapack-3.11.0-5_h47877c9_openblas.conda
-  build_number: 5
-  sha256: c723b6599fcd4c6c75dee728359ef418307280fa3e2ee376e14e85e5bbdda053
-  md5: b38076eb5c8e40d0106beda6f95d7609
-  depends:
-  - libblas 3.11.0 5_h4a7cf45_openblas
+  run_exports:
+    strong:
+    - _openmp_mutex >=4.5
+  size: 603998
+  timestamp: 1784923730278
+- conda: https://conda.anaconda.org/conda-forge/linux-64/liblapack-3.11.0-8_h47877c9_openblas.conda
+  build_number: 8
+  sha256: 168e327d737059553e15cc6ec36d76b9bbb3931c2a7721555fd68b4c9348b247
+  md5: 809be8ba8712c77bc7d44c2d99390dc4
+  depends:
+  - libblas 3.11.0 8_h4a7cf45_openblas
   constrains:
-  - blas 2.305   openblas
-  - liblapacke 3.11.0   5*_openblas
-  - libcblas   3.11.0   5*_openblas
+  - blas 2.308   openblas
+  - libcblas   3.11.0   8*_openblas
+  - liblapacke 3.11.0   8*_openblas
   license: BSD-3-Clause
   license_family: BSD
-  size: 18200
-  timestamp: 1765818857876
-- conda: https://conda.anaconda.org/conda-forge/linux-64/liblzma-5.8.2-hb03c661_0.conda
-  sha256: 755c55ebab181d678c12e49cced893598f2bab22d582fbbf4d8b83c18be207eb
-  md5: c7c83eecbb72d88b940c249af56c8b17
+  run_exports:
+    weak:
+    - liblapack >=3.11.0,<3.12.0a0
+  size: 18790
+  timestamp: 1779859115086
+- conda: https://conda.anaconda.org/conda-forge/linux-64/liblzma-5.8.3-hb03c661_0.conda
+  sha256: ec30e52a3c1bf7d0425380a189d209a52baa03f22fb66dd3eb587acaa765bd6d
+  md5: b88d90cad08e6bc8ad540cb310a761fb
   depends:
   - __glibc >=2.17,<3.0.a0
   - libgcc >=14
   constrains:
-  - xz 5.8.2.*
+  - xz 5.8.3.*
   license: 0BSD
-  size: 113207
-  timestamp: 1768752626120
+  run_exports:
+    weak:
+    - liblzma >=5.8.3,<6.0a0
+  size: 113478
+  timestamp: 1775825492909
 - conda: https://conda.anaconda.org/conda-forge/linux-64/libmpdec-4.0.0-hb03c661_1.conda
   sha256: fe171ed5cf5959993d43ff72de7596e8ac2853e9021dec0344e583734f1e0843
   md5: 2c21e66f50753a083cbe6b80f38268fa
@@ -379,170 +337,395 @@ packages:
   - libgcc >=14
   license: BSD-2-Clause
   license_family: BSD
+  run_exports: {}
   size: 92400
   timestamp: 1769482286018
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libopenblas-0.3.30-pthreads_h94d23a6_4.conda
-  sha256: 199d79c237afb0d4780ccd2fbf829cea80743df60df4705202558675e07dd2c5
-  md5: be43915efc66345cccb3c310b6ed0374
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libopenblas-0.3.33-pthreads_h94d23a6_0.conda
+  sha256: 3d9aa85648e5e18a6d66db98b8c4317cc426721ad7a220aa86330d1ccedc8903
+  md5: 2d3278b721e40468295ca755c3b84070
   depends:
   - __glibc >=2.17,<3.0.a0
   - libgcc >=14
   - libgfortran
   - libgfortran5 >=14.3.0
   constrains:
-  - openblas >=0.3.30,<0.3.31.0a0
+  - openblas >=0.3.33,<0.3.34.0a0
   license: BSD-3-Clause
   license_family: BSD
-  size: 5927939
-  timestamp: 1763114673331
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libsodium-1.0.20-h4ab18f5_0.conda
-  sha256: 0105bd108f19ea8e6a78d2d994a6d4a8db16d19a41212070d2d1d48a63c34161
-  md5: a587892d3c13b6621a6091be690dbca2
+  run_exports:
+    weak:
+    - libopenblas >=0.3.33,<1.0a0
+  size: 5931919
+  timestamp: 1776993658641
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libsodium-1.0.22-h280c20c_1.conda
+  sha256: b677bbf1c339d894757c3dcfbb2f88649e499e4991d70ae09a1466da9a6c92d6
+  md5: 965e4d531b588b2e42f66fd8e48b056c
   depends:
-  - libgcc-ng >=12
+  - __glibc >=2.17,<3.0.a0
+  - libgcc >=14
   license: ISC
-  size: 205978
-  timestamp: 1716828628198
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libsqlite-3.51.2-hf4e2dac_0.conda
-  sha256: 04596fcee262a870e4b7c9807224680ff48d4d0cc0dac076a602503d3dc6d217
-  md5: da5be73701eecd0e8454423fd6ffcf30
+  run_exports:
+    weak:
+    - libsodium >=1.0.22,<1.0.23.0a0
+  size: 269272
+  timestamp: 1779163468406
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libsqlite-3.53.4-hf4e2dac_0.conda
+  sha256: 72023efc207fe681e26b65fc9d668062cf0b4f0eacf3431e6eb099b95c1f2efd
+  md5: df088a279cd5e6fd2790b4c196434da1
   depends:
   - __glibc >=2.17,<3.0.a0
-  - icu >=78.2,<79.0a0
+  - icu >=78.3,<79.0a0
   - libgcc >=14
-  - libzlib >=1.3.1,<2.0a0
+  - libzlib >=1.3.2,<2.0a0
   license: blessing
-  size: 942808
-  timestamp: 1768147973361
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libstdcxx-15.2.0-h934c35e_17.conda
-  sha256: 50c48cd3716a2e58e8e2e02edc78fef2d08fffe1e3b1ed40eb5f87e7e2d07889
-  md5: 24c2fe35fa45cd71214beba6f337c071
+  run_exports:
+    weak:
+    - libsqlite >=3.53.4,<4.0a0
+  size: 964200
+  timestamp: 1785016112246
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libstdcxx-15.2.0-h934c35e_20.conda
+  sha256: b5eadda8fb0df262f0d424c4a0c8c1c8d65d904ce622f07936c793ea1b8a39d9
+  md5: fbd3d5506b11b5cfc916b29263b6b6f7
   depends:
   - __glibc >=2.17,<3.0.a0
-  - libgcc 15.2.0 he0feb66_17
+  - libgcc 15.2.0 he0feb66_20
   constrains:
-  - libstdcxx-ng ==15.2.0=*_17
+  - libstdcxx-ng ==15.2.0=*_20
   license: GPL-3.0-only WITH GCC-exception-3.1
   license_family: GPL
-  size: 5852406
-  timestamp: 1770252584235
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libstdcxx-ng-15.2.0-hdf11a46_17.conda
-  sha256: ca3fb322dab3373946b1064da686ec076f5b1b9caf0a2823dad00d0b0f704928
-  md5: ea12f5a6bf12c88c06750d9803e1a570
+  run_exports: {}
+  size: 5857690
+  timestamp: 1784923825011
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libuuid-2.42.2-h5347b49_0.conda
+  sha256: 9b1bdce27a7e31f7d241aeecff67a1f3101d52a2b1e33ccc2cdf2613072bf81f
+  md5: 01bb81d12c957de066ea7362007df642
   depends:
-  - libstdcxx 15.2.0 h934c35e_17
-  license: GPL-3.0-only WITH GCC-exception-3.1
-  license_family: GPL
-  size: 27573
-  timestamp: 1770252638797
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libuuid-2.41.3-h5347b49_0.conda
-  sha256: 1a7539cfa7df00714e8943e18de0b06cceef6778e420a5ee3a2a145773758aee
-  md5: db409b7c1720428638e7c0d509d3e1b5
-  depends:
-  - __glibc >=2.17,<3.0.a0
   - libgcc >=14
+  - __glibc >=2.17,<3.0.a0
   license: BSD-3-Clause
   license_family: BSD
-  size: 40311
-  timestamp: 1766271528534
-- conda: https://conda.anaconda.org/conda-forge/linux-64/libzlib-1.3.1-hb9d3cd8_2.conda
-  sha256: d4bfe88d7cb447768e31650f06257995601f89076080e76df55e3112d4e47dc4
-  md5: edb0dca6bc32e4f4789199455a1dbeb8
+  run_exports:
+    weak:
+    - libuuid >=2.42.2,<3.0a0
+  size: 40017
+  timestamp: 1781625522462
+- conda: https://conda.anaconda.org/conda-forge/linux-64/libzlib-1.3.2-h25fd6f3_2.conda
+  sha256: 55044c403570f0dc26e6364de4dc5368e5f3fc7ff103e867c487e2b5ab2bcda9
+  md5: d87ff7921124eccd67248aa483c23fec
   depends:
   - __glibc >=2.17,<3.0.a0
-  - libgcc >=13
   constrains:
-  - zlib 1.3.1 *_2
+  - zlib 1.3.2 *_2
   license: Zlib
   license_family: Other
-  size: 60963
-  timestamp: 1727963148474
-- conda: https://conda.anaconda.org/conda-forge/noarch/markdown-it-py-4.0.0-pyhd8ed1ab_0.conda
-  sha256: 7b1da4b5c40385791dbc3cc85ceea9fad5da680a27d5d3cb8bfaa185e304a89e
-  md5: 5b5203189eb668f042ac2b0826244964
+  run_exports:
+    weak:
+    - libzlib >=1.3.2,<2.0a0
+  size: 63629
+  timestamp: 1774072609062
+- conda: https://conda.anaconda.org/conda-forge/linux-64/msgspec-0.21.1-py314h5bd0f2a_0.conda
+  sha256: 52565ceea81e801c59dcaeaf5a9c77fba2fade445e67e0864fda50d4b944e15b
+  md5: 4a8ea416a56e58f012e445f7af2bbcc8
   depends:
-  - mdurl >=0.1,<1
-  - python >=3.10
-  license: MIT
-  license_family: MIT
-  size: 64736
-  timestamp: 1754951288511
-- conda: https://conda.modular.com/max-nightly/linux-64/max-26.2.0.dev2026022205-3.14release.conda
-  sha256: abe9988bd7aa6881b07157512251af5b5e0b930e319030e7923effd6c7bbe8ce
-  md5: 724d343713de47a6210ba3c8127c0d3b
+  - __glibc >=2.17,<3.0.a0
+  - libgcc >=14
+  - python >=3.14,<3.15.0a0
+  - python_abi 3.14.* *_cp314
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports: {}
+  size: 220990
+  timestamp: 1776337508167
+- conda: https://conda.anaconda.org/conda-forge/linux-64/ncurses-6.6-hdb14827_0.conda
+  sha256: fc89f74bbe362fb29fa3c037697a89bec140b346a2469a90f7936d1d7ea4d8a3
+  md5: fc21868a1a5aacc937e7a18747acb8a5
   depends:
-  - numpy >=1.18
-  - typing-extensions >=4.12.2
-  - pyyaml >=6.0.1
-  - rich >=13.0.1
-  - python-gil
-  - max-core ==26.2.0.dev2026022205 release
+  - __glibc >=2.17,<3.0.a0
+  - libgcc >=14
+  license: X11 AND BSD-3-Clause
+  run_exports:
+    weak:
+    - ncurses >=6.6,<7.0a0
+  size: 918956
+  timestamp: 1777422145199
+- conda: https://conda.anaconda.org/conda-forge/linux-64/numpy-2.5.1-py314h2b28147_0.conda
+  sha256: fb665b7e30f9472c58098983f614598cfcf34e7a26b6c419648803095c390aa7
+  md5: 59044905d27ba41bfb280ed4692c15e2
+  depends:
+  - python
+  - libgcc >=14
+  - __glibc >=2.17,<3.0.a0
+  - libstdcxx >=14
+  - libblas >=3.9.0,<4.0a0
+  - liblapack >=3.9.0,<4.0a0
   - python_abi 3.14.* *_cp314
+  - libcblas >=3.9.0,<4.0a0
   constrains:
-  - click >=8.0.0
-  - exceptiongroup >=0.2.2
-  - gguf >=0.17.1
-  - hf-transfer >=0.1.9
-  - huggingface_hub >=0.28.0
-  - llguidance >=0.7.30
-  - jinja2 >=3.1.0
-  - pillow >=11.0.0
-  - psutil >=6.1.1
-  - pydantic-settings >=2.7.1
-  - pydantic
-  - pydantic-core
-  - requests >=2.32.3
-  - sentencepiece >=0.2.0
-  - taskgroup >=0.2.2
-  - tqdm >=4.67.1
-  - transformers >=4.57.0,<5.0.0
-  - uvicorn >=0.34.0
-  - uvloop >=0.21.0
-  - aiofiles >=24.1.0
-  - asgiref >=3.8.1
-  - fastapi >=0.115.3
-  - grpcio >=1.68.0
-  - httpx >=0.28.1,<0.29
-  - msgspec >=0.19.0
-  - opentelemetry-api >=1.29.0
-  - opentelemetry-exporter-otlp-proto-http >=1.27.0
-  - opentelemetry-exporter-prometheus >=0.50b0
-  - opentelemetry-sdk >=1.29.0,<1.36.0
-  - prometheus_client >=0.21.0
-  - protobuf >=6.33.5,<6.34.0
-  - pyinstrument >=5.0.1
-  - python-json-logger >=2.0.7
-  - pyzmq >=26.3.0
-  - regex >=2024.11.6
-  - scipy >=1.13.0
-  - sse-starlette >=2.1.2
-  - starlette >=0.47.2
-  license: LicenseRef-Modular-Proprietary
-  size: 6106368
-  timestamp: 1771738763384
-- conda: https://conda.modular.com/max-nightly/linux-64/max-core-26.2.0.dev2026022205-release.conda
-  sha256: 898982c0b96cb40d83c08d84bf7a7295a58387f8c0be9571055f6fbbd5e9389b
-  md5: a25753c21792034b93c21cf09b47031e
+  - numpy-base <0a0
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports:
+    weak:
+    - numpy >=1.25,<3
+  size: 9089255
+  timestamp: 1783206203765
+- conda: https://conda.anaconda.org/conda-forge/linux-64/openssl-3.6.3-h35e630c_0.conda
+  sha256: d48f5c22b9897c01e4dff3680f1f57ceb02711ab9c62f74339b080419dfad34b
+  md5: 79dd2074b5cd5c5c6b2930514a11e22d
   depends:
-  - mojo-compiler ==0.26.2.0.dev2026022205 release
-  license: LicenseRef-Modular-Proprietary
-  size: 131486199
-  timestamp: 1771738763382
-- conda: https://conda.modular.com/max-nightly/noarch/mblack-26.2.0.dev2026022205-release.conda
+  - __glibc >=2.17,<3.0.a0
+  - ca-certificates
+  - libgcc >=14
+  license: Apache-2.0
+  license_family: Apache
+  run_exports:
+    weak:
+    - openssl >=3.6.3,<4.0a0
+  size: 3159683
+  timestamp: 1781069855778
+- conda: https://conda.anaconda.org/conda-forge/linux-64/psutil-7.2.2-py314h0f05182_0.conda
+  sha256: f15574ed6c8c8ed8c15a0c5a00102b1efe8b867c0bd286b498cd98d95bd69ae5
+  md5: 4f225a966cfee267a79c5cb6382bd121
+  depends:
+  - python
+  - libgcc >=14
+  - __glibc >=2.17,<3.0.a0
+  - python_abi 3.14.* *_cp314
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports: {}
+  size: 231303
+  timestamp: 1769678156552
+- conda: https://conda.anaconda.org/conda-forge/linux-64/python-3.14.6-habeac84_101_cp314.conda
+  build_number: 101
+  sha256: ee8f2006e1724b1f2e9e0ccc5a7cfdcab973460faa2f63ac1f6e44fdad4c0344
+  md5: 78975a41cf3c525da654f17e35bfca9e
+  depends:
+  - __glibc >=2.17,<3.0.a0
+  - bzip2 >=1.0.8,<2.0a0
+  - ld_impl_linux-64 >=2.36.1
+  - libexpat >=2.8.1,<3.0a0
+  - libffi >=3.5.2,<3.6.0a0
+  - libgcc >=14
+  - liblzma >=5.8.3,<6.0a0
+  - libmpdec >=4.0.0,<5.0a0
+  - libsqlite >=3.53.3,<4.0a0
+  - libuuid >=2.42.2,<3.0a0
+  - libzlib >=1.3.2,<2.0a0
+  - ncurses >=6.6,<7.0a0
+  - openssl >=3.5.7,<4.0a0
+  - python_abi 3.14.* *_cp314
+  - readline >=8.3,<9.0a0
+  - tk >=8.6.13,<8.7.0a0
+  - tzdata
+  - zstd >=1.5.7,<1.6.0a0
+  license: Python-2.0
+  run_exports:
+    weak:
+    - python_abi 3.14.* *_cp314
+    noarch:
+    - python
+  size: 36869055
+  timestamp: 1784910110714
+  python_site_packages_path: lib/python3.14/site-packages
+- conda: https://conda.anaconda.org/conda-forge/linux-64/pyzmq-27.1.0-py312hda471dd_3.conda
   noarch: python
-  sha256: 1e044bdd68205536a6a8e2158240bb1e3fc2e6eee294fcd181a9957c0eb7ade6
-  md5: aaa44d262d7410c49ab18353820f185f
+  sha256: 970b2a1d12983d8d1cc05d914ad88a0b6ef1fa14038c9649aa834dd6ebee65d7
+  md5: acd216255e1370e9aeab5351b831f07c
+  depends:
+  - python
+  - libgcc >=14
+  - libstdcxx >=14
+  - __glibc >=2.17,<3.0.a0
+  - _python_abi3_support 1.*
+  - cpython >=3.12
+  - zeromq >=4.3.5,<4.4.0a0
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports: {}
+  size: 210896
+  timestamp: 1779483879367
+- conda: https://conda.anaconda.org/conda-forge/linux-64/readline-8.3-h853b02a_0.conda
+  sha256: 12ffde5a6f958e285aa22c191ca01bbd3d6e710aa852e00618fa6ddc59149002
+  md5: d7d95fc8287ea7bf33e0e7116d2b95ec
+  depends:
+  - __glibc >=2.17,<3.0.a0
+  - libgcc >=14
+  - ncurses >=6.5,<7.0a0
+  license: GPL-3.0-only
+  license_family: GPL
+  run_exports:
+    weak:
+    - readline >=8.3,<9.0a0
+  size: 345073
+  timestamp: 1765813471974
+- conda: https://conda.anaconda.org/conda-forge/linux-64/tk-8.6.13-noxft_hd70dff1_3.conda
+  build_number: 103
+  sha256: 43624eab22f5f29df7d6ffe914cf442f28fd559b55b290906255492826e636e8
+  md5: 48a1049e710857572fc2a832aa394d9f
+  depends:
+  - __glibc >=2.17,<3.0.a0
+  - libgcc >=14
+  - libzlib >=1.3.2,<2.0a0
+  constrains:
+  - xorg-libx11 >=1.8.13,<2.0a0
+  license: TCL
+  run_exports:
+    weak:
+    - tk >=8.6.13,<8.7.0a0
+  size: 3550916
+  timestamp: 1784229071544
+- conda: https://conda.anaconda.org/conda-forge/linux-64/tornado-6.5.7-py314h5bd0f2a_0.conda
+  sha256: bbb7056f7c5fd606df16ed73ee68687050de2c02fd69a3f69a1cb533a7ed2ae8
+  md5: 4a8e5889712641aabdf6695e292857fe
+  depends:
+  - __glibc >=2.17,<3.0.a0
+  - libgcc >=14
+  - python >=3.14,<3.15.0a0
+  - python_abi 3.14.* *_cp314
+  license: Apache-2.0
+  license_family: Apache
+  run_exports: {}
+  size: 918368
+  timestamp: 1781006801436
+- conda: https://conda.anaconda.org/conda-forge/linux-64/zeromq-4.3.5-h09e67af_11.conda
+  sha256: dc9f28dedcb5f35a127fad2d847674d2833369dd616d294e423b8997df31d8a8
+  md5: 96b08867e21d4694fa5c2c226e6581b0
+  depends:
+  - libgcc >=14
+  - __glibc >=2.17,<3.0.a0
+  - libstdcxx >=14
+  - krb5 >=1.22.2,<1.23.0a0
+  - libsodium >=1.0.22,<1.0.23.0a0
+  license: MPL-2.0
+  license_family: MOZILLA
+  run_exports:
+    weak:
+    - zeromq >=4.3.5,<4.4.0a0
+  size: 311184
+  timestamp: 1779123989774
+- conda: https://conda.anaconda.org/conda-forge/linux-64/zstd-1.5.7-hb78ec9c_6.conda
+  sha256: 68f0206ca6e98fea941e5717cec780ed2873ffabc0e1ed34428c061e2c6268c7
+  md5: 4a13eeac0b5c8e5b8ab496e6c4ddd829
+  depends:
+  - __glibc >=2.17,<3.0.a0
+  - libzlib >=1.3.1,<2.0a0
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports:
+    weak:
+    - zstd >=1.5.7,<1.6.0a0
+  size: 601375
+  timestamp: 1764777111296
+- conda: https://conda.anaconda.org/conda-forge/noarch/_python_abi3_support-1.0-hd8ed1ab_3.conda
+  sha256: 2a7204314663eeda5dec482a956f0e2eaf289bd5b9953eaaaad0e81aa64638f2
+  md5: 3845f3d75991bae0fb90884662f4327c
+  depends:
+  - cpython
+  - python-gil
+  license: MIT
+  license_family: MIT
+  run_exports: {}
+  size: 8144
+  timestamp: 1784221492234
+- conda: https://conda.anaconda.org/conda-forge/noarch/ca-certificates-2026.7.22-hbd8a1cb_0.conda
+  sha256: 0a0544cf95f64394fe4959286f5c71f5444ad58feb0602e53becb27448d24da6
+  md5: 0f51e2391ade309db462a55611263e9c
+  depends:
+  - __unix
+  license: ISC
+  run_exports: {}
+  size: 131780
+  timestamp: 1784754889428
+- conda: https://conda.anaconda.org/conda-forge/noarch/click-8.4.2-pyhc90fa1f_0.conda
+  sha256: ccc4787f511964f9a1f2d2d2859c91c5d571fb60f7f09d4c4e092c9b7a94e671
+  md5: 2c4bd6aeb90bb157456841c3270a0d92
   depends:
+  - __unix
+  - python
+  - python >=3.10
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports: {}
+  size: 107155
+  timestamp: 1783085363526
+- conda: https://conda.anaconda.org/conda-forge/noarch/cpython-3.14.6-py314hd8ed1ab_101.conda
+  noarch: generic
+  sha256: 436618a5a090c9f7ade0c5f883e8602a130b3991bc88b06badd6f805d7dbff00
+  md5: 424c465894c8af725105fe6ad74f6aec
+  depends:
+  - python >=3.14,<3.15.0a0
+  - python_abi * *_cp314
+  license: Python-2.0
+  run_exports: {}
+  size: 49508
+  timestamp: 1784909547134
+- conda: https://conda.anaconda.org/conda-forge/noarch/exceptiongroup-1.3.1-pyhd8ed1ab_0.conda
+  sha256: ee6cf346d017d954255bbcbdb424cddea4d14e4ed7e9813e429db1d795d01144
+  md5: 8e662bd460bda79b1ea39194e3c4c9ab
+  depends:
+  - python >=3.10
+  - typing_extensions >=4.6.0
+  license: MIT and PSF-2.0
+  run_exports: {}
+  size: 21333
+  timestamp: 1763918099466
+- conda: https://conda.anaconda.org/conda-forge/noarch/importlib-metadata-9.0.0-pyhcf101f3_0.conda
+  sha256: 43e2a5497cad1598ff88a3e69f69bc88b7b8f141fa63c60eab5db296317318b8
+  md5: ffc17e785d64e12fc311af9184221839
+  depends:
+  - python >=3.10
+  - zipp >=3.20
+  - python
+  license: Apache-2.0
+  license_family: APACHE
+  run_exports: {}
+  size: 34766
+  timestamp: 1779714582554
+- conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_client-8.6.3-pyhd8ed1ab_1.conda
+  sha256: 19d8bd5bb2fde910ec59e081eeb59529491995ce0d653a5209366611023a0b3a
+  md5: 4ebae00eae9705b0c3d6d1018a81d047
+  depends:
+  - importlib-metadata >=4.8.3
+  - jupyter_core >=4.12,!=5.0.*
+  - python >=3.9
+  - python-dateutil >=2.8.2
+  - pyzmq >=23.0
+  - tornado >=6.2
+  - traitlets >=5.3
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports: {}
+  size: 106342
+  timestamp: 1733441040958
+- conda: https://conda.anaconda.org/conda-forge/noarch/jupyter_core-5.9.1-pyhc90fa1f_0.conda
+  sha256: 1d34b80e5bfcd5323f104dbf99a2aafc0e5d823019d626d0dce5d3d356a2a52a
+  md5: b38fe4e78ee75def7e599843ef4c1ab0
+  depends:
+  - __unix
+  - python
+  - platformdirs >=2.5
+  - python >=3.10
+  - traitlets >=5.3
+  - python
+  constrains:
+  - pywin32 >=300
+  license: BSD-3-Clause
+  license_family: BSD
+  run_exports: {}
+  size: 65503
+  timestamp: 1760643864586
+- conda: https://conda.anaconda.org/conda-forge/noarch/markdown-it-py-4.2.0-pyhd8ed1ab_0.conda
+  sha256: 0c4c35376fe920714390d46e4b8d31c876d65f18e1655899e0763ec25f2a902f
+  md5: 6d03368f2b2b0a5fb6839df53b2eb5e0
+  depends:
+  - mdurl >=0.1,<1
   - python >=3.10
-  - click >=8.0.0
-  - mypy_extensions >=0.4.3
-  - packaging >=22.0
-  - pathspec >=0.9.0
-  - platformdirs >=2
-  - tomli >=1.1.0
   license: MIT
-  size: 135129
-  timestamp: 1771738763383
+  license_family: MIT
+  run_exports: {}
+  size: 69017
+  timestamp: 1778169663339
 - conda: https://conda.anaconda.org/conda-forge/noarch/mdurl-0.1.2-pyhd8ed1ab_1.conda
   sha256: 78c1bbe1723449c52b7a9df1af2ee5f005209f67e40b6e1d3c7619127c43b1c7
   md5: 592132998493b3ff25fd7479396e8351
@@ -550,36 +733,9 @@ packages:
   - python >=3.9
   license: MIT
   license_family: MIT
+  run_exports: {}
   size: 14465
   timestamp: 1733255681319
-- conda: https://conda.modular.com/max-nightly/linux-64/mojo-0.26.2.0.dev2026022205-release.conda
-  sha256: 0d8ca62be9a970f4382c99549fb88c39c7dc28092ee812f03f1639990d4948b2
-  md5: 1275bc6aedb6410368dd1bf6d98f6f90
-  depends:
-  - python >=3.10
-  - mojo-compiler ==0.26.2.0.dev2026022205 release
-  - mblack ==26.2.0.dev2026022205 release
-  - jupyter_client >=8.6.2,<8.7
-  license: LicenseRef-Modular-Proprietary
-  size: 88491769
-  timestamp: 1771738763383
-- conda: https://conda.modular.com/max-nightly/linux-64/mojo-compiler-0.26.2.0.dev2026022205-release.conda
-  sha256: 173dd0f2401b3342d7cf59c29745a84f2c1168976a6f102318631b81c370c6b4
-  md5: c8db27767e0be9250b72e3414602ff37
-  depends:
-  - mojo-python ==0.26.2.0.dev2026022205 release
-  license: LicenseRef-Modular-Proprietary
-  size: 87474357
-  timestamp: 1771738763382
-- conda: https://conda.modular.com/max-nightly/noarch/mojo-python-0.26.2.0.dev2026022205-release.conda
-  noarch: python
-  sha256: 71199f2ae34cfd0cfcce57613448d3d54e000151908ca9564ccb8e1b6beb4ddd
-  md5: 7742228841ea05a513e667fbf57fdfb2
-  depends:
-  - python
-  license: LicenseRef-Modular-Proprietary
-  size: 676288
-  timestamp: 1771738763380
 - conda: https://conda.anaconda.org/conda-forge/noarch/mypy_extensions-1.1.0-pyha770c72_0.conda
   sha256: 6ed158e4e5dd8f6a10ad9e525631e35cee8557718f83de7a4e3966b1f772c4b1
   md5: e9c622e0d00fa24a6292279af3ab6d06
@@ -587,110 +743,51 @@ packages:
   - python >=3.9
   license: MIT
   license_family: MIT
+  run_exports: {}
   size: 11766
   timestamp: 1745776666688
-- conda: https://conda.anaconda.org/conda-forge/linux-64/ncurses-6.5-h2d0b736_3.conda
-  sha256: 3fde293232fa3fca98635e1167de6b7c7fda83caf24b9d6c91ec9eefb4f4d586
-  md5: 47e340acb35de30501a76c7c799c41d7
-  depends:
-  - __glibc >=2.17,<3.0.a0
-  - libgcc >=13
-  license: X11 AND BSD-3-Clause
-  size: 891641
-  timestamp: 1738195959188
-- conda: https://conda.anaconda.org/conda-forge/linux-64/numpy-2.4.2-py314h2b28147_1.conda
-  sha256: 1d8377c8001c15ed12c2713b723213474b435706ab9d34ede69795d64af9e94d
-  md5: 4ea6b620fdf24a1a0bc4f1c7134dfafb
-  depends:
-  - python
-  - libstdcxx >=14
-  - libgcc >=14
-  - __glibc >=2.17,<3.0.a0
-  - libcblas >=3.9.0,<4.0a0
-  - python_abi 3.14.* *_cp314
-  - libblas >=3.9.0,<4.0a0
-  - liblapack >=3.9.0,<4.0a0
-  constrains:
-  - numpy-base <0a0
-  license: BSD-3-Clause
-  license_family: BSD
-  size: 8926994
-  timestamp: 1770098474394
-- conda: https://conda.anaconda.org/conda-forge/linux-64/openssl-3.6.1-h35e630c_1.conda
-  sha256: 44c877f8af015332a5d12f5ff0fb20ca32f896526a7d0cdb30c769df1144fb5c
-  md5: f61eb8cd60ff9057122a3d338b99c00f
-  depends:
-  - __glibc >=2.17,<3.0.a0
-  - ca-certificates
-  - libgcc >=14
-  license: Apache-2.0
-  license_family: Apache
-  size: 3164551
-  timestamp: 1769555830639
-- conda: https://conda.anaconda.org/conda-forge/noarch/packaging-26.0-pyhcf101f3_0.conda
-  sha256: c1fc0f953048f743385d31c468b4a678b3ad20caffdeaa94bed85ba63049fd58
-  md5: b76541e68fea4d511b1ac46a28dcd2c6
+- conda: https://conda.anaconda.org/conda-forge/noarch/packaging-26.2-pyhc364b38_0.conda
+  sha256: 3906abfb6511a3bb309e39b9b1b7bc38f50a723971de2395489fd1f379255890
+  md5: 4c06a92e74452cfa53623a81592e8934
   depends:
   - python >=3.8
   - python
   license: Apache-2.0
   license_family: APACHE
-  size: 72010
-  timestamp: 1769093650580
-- conda: https://conda.anaconda.org/conda-forge/noarch/pathspec-1.0.4-pyhd8ed1ab_0.conda
-  sha256: 29ea20d0faf20374fcd61c25f6d32fb8e9a2c786a7f1473a0c3ead359470fbe1
-  md5: 2908273ac396d2cd210a8127f5f1c0d6
+  run_exports: {}
+  size: 91574
+  timestamp: 1777103621679
+- conda: https://conda.anaconda.org/conda-forge/noarch/pathspec-1.1.1-pyhd8ed1ab_0.conda
+  sha256: 6eaee417d33f298db79bc7185ab1208604c0e6cf51dade34cd513c6f9db9c6f3
+  md5: 11adc78451c998c0fd162584abfa3559
   depends:
   - python >=3.10
   license: MPL-2.0
   license_family: MOZILLA
-  size: 53739
-  timestamp: 1769677743677
-- conda: https://conda.anaconda.org/conda-forge/noarch/platformdirs-4.7.0-pyhcf101f3_0.conda
-  sha256: 5b14e300187919e783b24507dbeaeee891d97a3e9a9c80b5dcd3073753bac69c
-  md5: 2157d0900a4bc2e9a0ba3cccb8497e8c
+  run_exports: {}
+  size: 56559
+  timestamp: 1777271601895
+- conda: https://conda.anaconda.org/conda-forge/noarch/platformdirs-4.11.0-pyhcf101f3_0.conda
+  sha256: a4b8c7d3b3703d10f93187986d59aec09b22033978945845899beb7f849a7be6
+  md5: 1fadaa6dd1d03d062075f84157ac2cc7
   depends:
   - python >=3.10
   - python
   license: MIT
-  size: 24091
-  timestamp: 1770990257318
-- conda: https://conda.anaconda.org/conda-forge/noarch/pygments-2.19.2-pyhd8ed1ab_0.conda
-  sha256: 5577623b9f6685ece2697c6eb7511b4c9ac5fb607c9babc2646c811b428fd46a
-  md5: 6b6ece66ebcae2d5f326c77ef2c5a066
+  license_family: MIT
+  run_exports: {}
+  size: 26632
+  timestamp: 1784661349391
+- conda: https://conda.anaconda.org/conda-forge/noarch/pygments-2.20.0-pyhd8ed1ab_0.conda
+  sha256: cf70b2f5ad9ae472b71235e5c8a736c9316df3705746de419b59d442e8348e86
+  md5: 16c18772b340887160c79a6acc022db0
   depends:
-  - python >=3.9
+  - python >=3.10
   license: BSD-2-Clause
   license_family: BSD
-  size: 889287
-  timestamp: 1750615908735
-- conda: https://conda.anaconda.org/conda-forge/linux-64/python-3.14.3-h32b2ec7_101_cp314.conda
-  build_number: 101
-  sha256: cb0628c5f1732f889f53a877484da98f5a0e0f47326622671396fb4f2b0cd6bd
-  md5: c014ad06e60441661737121d3eae8a60
-  depends:
-  - __glibc >=2.17,<3.0.a0
-  - bzip2 >=1.0.8,<2.0a0
-  - ld_impl_linux-64 >=2.36.1
-  - libexpat >=2.7.3,<3.0a0
-  - libffi >=3.5.2,<3.6.0a0
-  - libgcc >=14
-  - liblzma >=5.8.2,<6.0a0
-  - libmpdec >=4.0.0,<5.0a0
-  - libsqlite >=3.51.2,<4.0a0
-  - libuuid >=2.41.3,<3.0a0
-  - libzlib >=1.3.1,<2.0a0
-  - ncurses >=6.5,<7.0a0
-  - openssl >=3.5.5,<4.0a0
-  - python_abi 3.14.* *_cp314
-  - readline >=8.3,<9.0a0
-  - tk >=8.6.13,<8.7.0a0
-  - tzdata
-  - zstd >=1.5.7,<1.6.0a0
-  license: Python-2.0
-  size: 36702440
-  timestamp: 1770675584356
-  python_site_packages_path: lib/python3.14/site-packages
+  run_exports: {}
+  size: 893031
+  timestamp: 1774796815820
 - conda: https://conda.anaconda.org/conda-forge/noarch/python-dateutil-2.9.0.post0-pyhe01879c_2.conda
   sha256: d6a17ece93bbd5139e02d2bd7dbfa80bee1a4261dced63f65f679121686bf664
   md5: 5b8d21249ff20967101ffa321cab24e8
@@ -700,17 +797,19 @@ packages:
   - python
   license: Apache-2.0
   license_family: APACHE
+  run_exports: {}
   size: 233310
   timestamp: 1751104122689
-- conda: https://conda.anaconda.org/conda-forge/noarch/python-gil-3.14.3-h4df99d1_101.conda
-  sha256: 233aebd94c704ac112afefbb29cf4170b7bc606e22958906f2672081bc50638a
-  md5: 235765e4ea0d0301c75965985163b5a1
+- conda: https://conda.anaconda.org/conda-forge/noarch/python-gil-3.14.6-h4df99d1_101.conda
+  sha256: f96f61b33fe7d0f599ba5e23d9e5231fad9c8a37a1c141dfa8edbd0ba78de9e3
+  md5: 7f742295acd62ee0688e7e6924b71e67
   depends:
-  - cpython 3.14.3.*
+  - cpython 3.14.6.*
   - python_abi * *_cp314
   license: Python-2.0
-  size: 50062
-  timestamp: 1770674497152
+  run_exports: {}
+  size: 49484
+  timestamp: 1784909578801
 - conda: https://conda.anaconda.org/conda-forge/noarch/python_abi-3.14-8_cp314.conda
   build_number: 8
   sha256: ad6d2e9ac39751cc0529dd1566a26751a0bf2542adb0c232533d32e176e21db5
@@ -719,51 +818,12 @@ packages:
   - python 3.14.* *_cp314
   license: BSD-3-Clause
   license_family: BSD
+  run_exports: {}
   size: 6989
   timestamp: 1752805904792
-- conda: https://conda.anaconda.org/conda-forge/linux-64/pyyaml-6.0.3-py314h67df5f8_1.conda
-  sha256: b318fb070c7a1f89980ef124b80a0b5ccf3928143708a85e0053cde0169c699d
-  md5: 2035f68f96be30dc60a5dfd7452c7941
-  depends:
-  - __glibc >=2.17,<3.0.a0
-  - libgcc >=14
-  - python >=3.14,<3.15.0a0
-  - python_abi 3.14.* *_cp314
-  - yaml >=0.2.5,<0.3.0a0
-  license: MIT
-  license_family: MIT
-  size: 202391
-  timestamp: 1770223462836
-- conda: https://conda.anaconda.org/conda-forge/linux-64/pyzmq-27.1.0-py312hfb55c3c_0.conda
-  noarch: python
-  sha256: a00a41b66c12d9c60e66b391e9a4832b7e28743348cf4b48b410b91927cd7819
-  md5: 3399d43f564c905250c1aea268ebb935
-  depends:
-  - python
-  - __glibc >=2.17,<3.0.a0
-  - libstdcxx >=14
-  - libgcc >=14
-  - _python_abi3_support 1.*
-  - cpython >=3.12
-  - zeromq >=4.3.5,<4.4.0a0
-  license: BSD-3-Clause
-  license_family: BSD
-  size: 212218
-  timestamp: 1757387023399
-- conda: https://conda.anaconda.org/conda-forge/linux-64/readline-8.3-h853b02a_0.conda
-  sha256: 12ffde5a6f958e285aa22c191ca01bbd3d6e710aa852e00618fa6ddc59149002
-  md5: d7d95fc8287ea7bf33e0e7116d2b95ec
-  depends:
-  - __glibc >=2.17,<3.0.a0
-  - libgcc >=14
-  - ncurses >=6.5,<7.0a0
-  license: GPL-3.0-only
-  license_family: GPL
-  size: 345073
-  timestamp: 1765813471974
-- conda: https://conda.anaconda.org/conda-forge/noarch/rich-14.3.2-pyhcf101f3_0.conda
-  sha256: ed17985cec5a0540002c6cabe67848f7cc17e5f4019c0e2a40534e9b7c0b38de
-  md5: 33950a076fd589a7655c6888cc3d2b34
+- conda: https://conda.anaconda.org/conda-forge/noarch/rich-15.0.0-pyhcf101f3_0.conda
+  sha256: 3d6ba2c0fcdac3196ba2f0615b4104e532525ffa1335b50a2878be5ff488814a
+  md5: 0242025a3c804966bf71aa04eee82f66
   depends:
   - markdown-it-py >=2.2.0
   - pygments >=2.13.0,<3.0.0
@@ -772,8 +832,9 @@ packages:
   - python
   license: MIT
   license_family: MIT
-  size: 208269
-  timestamp: 1769971520792
+  run_exports: {}
+  size: 208577
+  timestamp: 1775991661559
 - conda: https://conda.anaconda.org/conda-forge/noarch/six-1.17.0-pyhe01879c_1.conda
   sha256: 458227f759d5e3fcec5d9b7acce54e10c9e1f4f4b7ec978f3bfd54ce4ee9853d
   md5: 3339e3b65d58accf4ca4fb8748ab16b3
@@ -782,118 +843,154 @@ packages:
   - python
   license: MIT
   license_family: MIT
+  run_exports: {}
   size: 18455
   timestamp: 1753199211006
-- conda: https://conda.anaconda.org/conda-forge/linux-64/tk-8.6.13-noxft_h366c992_103.conda
-  sha256: cafeec44494f842ffeca27e9c8b0c27ed714f93ac77ddadc6aaf726b5554ebac
-  md5: cffd3bdd58090148f4cfcd831f4b26ab
+- conda: https://conda.anaconda.org/conda-forge/noarch/taskgroup-0.2.2-pyhd8ed1ab_0.conda
+  sha256: 6f8db6da8de445930de55b708e6a5d3ab5f076bc14a39578db0190b2a9b8e437
+  md5: 9fa69537fb68a095fbac139210575bad
   depends:
-  - __glibc >=2.17,<3.0.a0
-  - libgcc >=14
-  - libzlib >=1.3.1,<2.0a0
-  constrains:
-  - xorg-libx11 >=1.8.12,<2.0a0
-  license: TCL
-  license_family: BSD
-  size: 3301196
-  timestamp: 1769460227866
-- conda: https://conda.anaconda.org/conda-forge/noarch/tomli-2.4.0-pyhcf101f3_0.conda
-  sha256: 62940c563de45790ba0f076b9f2085a842a65662268b02dd136a8e9b1eaf47a8
-  md5: 72e780e9aa2d0a3295f59b1874e3768b
+  - exceptiongroup
+  - python >=3.9
+  - typing_extensions >=4.12.2,<5
+  license: MIT
+  license_family: MIT
+  run_exports: {}
+  size: 17330
+  timestamp: 1736003478648
+- conda: https://conda.anaconda.org/conda-forge/noarch/tomli-2.4.1-pyhcf101f3_0.conda
+  sha256: 91cafdb64268e43e0e10d30bd1bef5af392e69f00edd34dfaf909f69ab2da6bd
+  md5: b5325cf06a000c5b14970462ff5e4d58
   depends:
   - python >=3.10
   - python
   license: MIT
   license_family: MIT
-  size: 21453
-  timestamp: 1768146676791
-- conda: https://conda.anaconda.org/conda-forge/linux-64/tornado-6.5.3-py314h5bd0f2a_0.conda
-  sha256: b8f9f9ae508d79c9c697eb01b6a8d2ed4bc1899370f44aa6497c8abbd15988ea
-  md5: e35f08043f54d26a1be93fdbf90d30c3
-  depends:
-  - __glibc >=2.17,<3.0.a0
-  - libgcc >=14
-  - python >=3.14,<3.15.0a0
-  - python_abi 3.14.* *_cp314
-  license: Apache-2.0
-  license_family: Apache
-  size: 905436
-  timestamp: 1765458949518
-- conda: https://conda.anaconda.org/conda-forge/noarch/traitlets-5.14.3-pyhd8ed1ab_1.conda
-  sha256: f39a5620c6e8e9e98357507262a7869de2ae8cc07da8b7f84e517c9fd6c2b959
-  md5: 019a7385be9af33791c989871317e1ed
+  run_exports: {}
+  size: 21561
+  timestamp: 1774492402955
+- conda: https://conda.anaconda.org/conda-forge/noarch/traitlets-5.15.1-pyhcf101f3_0.conda
+  sha256: b89a823edf524956b94a2a4db974866e4501f05c68976eff458c5dcf07f88431
+  md5: 37e3be7b6e2977d37b8fa5da229f5dc0
   depends:
-  - python >=3.9
+  - python >=3.10
+  - python
   license: BSD-3-Clause
   license_family: BSD
-  size: 110051
-  timestamp: 1733367480074
-- conda: https://conda.anaconda.org/conda-forge/noarch/typing-extensions-4.15.0-h396c80c_0.conda
-  sha256: 7c2df5721c742c2a47b2c8f960e718c930031663ac1174da67c1ed5999f7938c
-  md5: edd329d7d3a4ab45dcf905899a7a6115
-  depends:
-  - typing_extensions ==4.15.0 pyhcf101f3_0
+  run_exports: {}
+  size: 115158
+  timestamp: 1780507822178
+- conda: https://conda.anaconda.org/conda-forge/noarch/typing-extensions-4.16.0-h69aa097_0.conda
+  sha256: b141933ece3518f6d7b75dfb59451e2f26b405a44c18e2518a83e9a02e09315c
+  md5: c680b5747e8c4c8f23dca0bb7042a8fc
+  depends:
+  - typing_extensions ==4.16.0 pyhcf101f3_0
   license: PSF-2.0
   license_family: PSF
-  size: 91383
-  timestamp: 1756220668932
-- conda: https://conda.anaconda.org/conda-forge/noarch/typing_extensions-4.15.0-pyhcf101f3_0.conda
-  sha256: 032271135bca55aeb156cee361c81350c6f3fb203f57d024d7e5a1fc9ef18731
-  md5: 0caa1af407ecff61170c9437a808404d
+  run_exports: {}
+  size: 94080
+  timestamp: 1783002732887
+- conda: https://conda.anaconda.org/conda-forge/noarch/typing_extensions-4.16.0-pyhcf101f3_0.conda
+  sha256: 2d888f90af0686044882c74193ec80a90ec1943145d94a7b1b048958acda1848
+  md5: c70ad746c22219b9700931707482992c
   depends:
   - python >=3.10
   - python
   license: PSF-2.0
   license_family: PSF
-  size: 51692
-  timestamp: 1756220668932
-- conda: https://conda.anaconda.org/conda-forge/noarch/tzdata-2025c-hc9c84f9_1.conda
-  sha256: 1d30098909076af33a35017eed6f2953af1c769e273a0626a04722ac4acaba3c
-  md5: ad659d0a2b3e47e38d829aa8cad2d610
+  run_exports: {}
+  size: 52631
+  timestamp: 1783002732887
+- conda: https://conda.anaconda.org/conda-forge/noarch/tzdata-2026c-h151e31d_0.conda
+  sha256: b928c30ddcb0e3f544c6eade8352737e6e610e263276b90232db6a578ef899d8
+  md5: fcb489df604d100968b737f2cb6076c6
   license: LicenseRef-Public-Domain
-  size: 119135
-  timestamp: 1767016325805
-- conda: https://conda.anaconda.org/conda-forge/linux-64/yaml-0.2.5-h280c20c_3.conda
-  sha256: 6d9ea2f731e284e9316d95fa61869fe7bbba33df7929f82693c121022810f4ad
-  md5: a77f85f77be52ff59391544bfe73390a
+  run_exports: {}
+  size: 118849
+  timestamp: 1784250406640
+- conda: https://conda.anaconda.org/conda-forge/noarch/zipp-4.1.0-pyhcf101f3_0.conda
+  sha256: 210bd31c22bb88f5e2a167df24c95bb5f152b2ada7502f9b8c49d1f5366db423
+  md5: ba3dcdc8584155c97c648ae9c044b7a3
   depends:
-  - libgcc >=14
-  - __glibc >=2.17,<3.0.a0
+  - python >=3.10
+  - python
   license: MIT
   license_family: MIT
-  size: 85189
-  timestamp: 1753484064210
-- conda: https://conda.anaconda.org/conda-forge/linux-64/zeromq-4.3.5-h387f397_9.conda
-  sha256: 47cfe31255b91b4a6fa0e9dbaf26baa60ac97e033402dbc8b90ba5fee5ffe184
-  md5: 8035e5b54c08429354d5d64027041cad
+  run_exports: {}
+  size: 24190
+  timestamp: 1779159948016
+- conda: https://conda.modular.com/max-nightly/linux-64/max-26.5.0.dev2026072505-3.14release.conda
+  sha256: 735cc65e1938c0c0e5182d96110a144239782b5d954ebddbad4dfcb0a246637b
+  md5: ffe92be5964a56b949a8c782a3f0a98d
   depends:
-  - libstdcxx >=14
-  - libgcc >=14
-  - __glibc >=2.17,<3.0.a0
-  - libgcc >=14
-  - libsodium >=1.0.20,<1.0.21.0a0
-  - krb5 >=1.21.3,<1.22.0a0
-  license: MPL-2.0
-  license_family: MOZILLA
-  size: 310648
-  timestamp: 1757370847287
-- conda: https://conda.anaconda.org/conda-forge/noarch/zipp-3.23.0-pyhcf101f3_1.conda
-  sha256: b4533f7d9efc976511a73ef7d4a2473406d7f4c750884be8e8620b0ce70f4dae
-  md5: 30cd29cb87d819caead4d55184c1d115
+  - click >=8.0.0
+  - exceptiongroup >=0.2.2
+  - msgspec >=0.19.0
+  - numpy >=1.18
+  - psutil >=7.0.0
+  - rich >=13.0.1
+  - taskgroup >=0.2.2
+  - typing-extensions >=4.12.2
+  - python 3.14.*
+  - python-gil
+  - max-core ==26.5.0.dev2026072505
+  license: LicenseRef-Modular-Proprietary
+  run_exports: {}
+  size: 18225480
+  timestamp: 1784959682216
+- conda: https://conda.modular.com/max-nightly/linux-64/max-core-26.5.0.dev2026072505-release.conda
+  sha256: a2caecf6f29447375ce1bb42979d998c13acf5f48f5b16c7dc6166825c005cf0
+  md5: 20809390abf8b0e993718841eb16813b
+  depends:
+  - mojo-compiler ==1.0.0b3.dev2026072505
+  license: LicenseRef-Modular-Proprietary
+  run_exports: {}
+  size: 103408905
+  timestamp: 1784959675208
+- conda: https://conda.modular.com/max-nightly/linux-64/mojo-1.0.0b3.dev2026072505-release.conda
+  sha256: 9fa11e681371d1994ffbe66d0f7e1502e7b22381c5ca8d5f844b77a335c8e19b
+  md5: 6a0c4ee8b9b53aba98644702a4c2c682
   depends:
   - python >=3.10
-  - python
-  license: MIT
-  license_family: MIT
-  size: 24194
-  timestamp: 1764460141901
-- conda: https://conda.anaconda.org/conda-forge/linux-64/zstd-1.5.7-hb78ec9c_6.conda
-  sha256: 68f0206ca6e98fea941e5717cec780ed2873ffabc0e1ed34428c061e2c6268c7
-  md5: 4a13eeac0b5c8e5b8ab496e6c4ddd829
+  - mojo-compiler ==1.0.0b3.dev2026072505
+  - mblack ==26.5.0.dev2026072505
+  - jupyter_client >=8.6.2,<8.7
+  license: LicenseRef-Modular-Proprietary
+  run_exports: {}
+  size: 114479370
+  timestamp: 1784959545074
+- conda: https://conda.modular.com/max-nightly/linux-64/mojo-compiler-1.0.0b3.dev2026072505-release.conda
+  sha256: 307aae574f073cf914c9efe3cfadeeb2f3992c3a426154731e934b81f7b2f725
+  md5: b9d5de41a9c940f47090818f8d9d0c18
+  depends:
+  - mojo-python ==1.0.0b3.dev2026072505
+  license: LicenseRef-Modular-Proprietary
+  run_exports: {}
+  size: 80560693
+  timestamp: 1784959541118
+- conda: https://conda.modular.com/max-nightly/noarch/mblack-26.5.0.dev2026072505-release.conda
+  noarch: python
+  sha256: a66e15918fe18928e4a35c08e16f0517c25d5bc10f836a2240c7cbf9788f9501
+  md5: 19378925651cd95d631a7f580cb2846d
   depends:
-  - __glibc >=2.17,<3.0.a0
-  - libzlib >=1.3.1,<2.0a0
-  license: BSD-3-Clause
-  license_family: BSD
-  size: 601375
-  timestamp: 1764777111296
+  - python >=3.10
+  - click >=8.0.0
+  - mypy_extensions >=0.4.3
+  - packaging >=22.0
+  - pathspec >=0.9.0
+  - platformdirs >=2
+  - tomli >=1.1.0
+  license: LicenseRef-Modular-Proprietary
+  run_exports: {}
+  size: 136731
+  timestamp: 1784959402726
+- conda: https://conda.modular.com/max-nightly/noarch/mojo-python-1.0.0b3.dev2026072505-release.conda
+  noarch: python
+  sha256: 9d90f9d14440c864e96c331fb54e5a2be5fc724bdcdec0adf9ddd52d1dabdcb9
+  md5: f5e33c79f95b3cb3bd7e7d4bfe227eb2
+  depends:
+  - python >=3.10
+  license: LicenseRef-Modular-Proprietary
+  run_exports: {}
+  size: 24107
+  timestamp: 1784959402688
diff --git a/benchmark/pixi.toml b/mojo/pixi.toml
similarity index 67%
rename from benchmark/pixi.toml
rename to mojo/pixi.toml
index 46776c37af..66bc00e163 100644
--- a/benchmark/pixi.toml
+++ b/mojo/pixi.toml
@@ -1,12 +1,12 @@
 [workspace]
 authors = ["maxchisto <max.chisto@gmail.com>"]
 channels = ["https://conda.modular.com/max-nightly", "conda-forge"]
-name = "benchmark"
+name = "mojo"
 platforms = ["linux-64"]
 version = "0.1.0"
 
 [tasks]
 
 [dependencies]
-mojo = ">=0.26.2.0.dev2026022205,<0.27"
-max = ">=26.2.0.dev2026022205,<27"
+mojo = ">=1.0.0b3.dev2026072505,<2"
+max = ">=26.5.0.dev2026072505"
diff --git a/mojo/sigmoid_build_instructions.md b/mojo/sigmoid_build_instructions.md
new file mode 100644
index 0000000000..dc107c6639
--- /dev/null
+++ b/mojo/sigmoid_build_instructions.md
@@ -0,0 +1,39 @@
+# Build Instructions
+
+## 1. Compile the Mojo Module (`libsigmoid_mojo.so`)
+
+The Mojo code provides the core processing logic and kernels.
+
+```bash
+cd ~/code/darktable/mojo
+make build 
+```
+
+The resulting `libsigmoid_mojo.so` contains the following exported symbols used by the C bridge:
+- `sigmoid_mojo_init`
+- `sigmoid_mojo_destroy`
+- `sigmoid_mojo_rgb_ratio`
+- `sigmoid_mojo_per_channel`
+
+## 2. Compile the C-Bridge Plugin (`libsigmoid.so`)
+
+The C part (`src/iop/sigmoid.c`) handles the darktable user interface and parameter management. It loads the Mojo shared library at runtime.
+
+```bash
+./build_sigmoid_iop.sh
+```
+
+
+## 3. Installation
+
+Both shared libraries must be placed in the darktable plugins directory so that darktable can find the plugin and the plugin can find the Mojo library.
+
+```bash
+# Copy the Mojo library
+sudo cp libsigmoid_mojo.so /usr/lib/darktable/plugins/
+
+# Copy the C bridge plugin
+sudo cp libsigmoid.so /usr/lib/darktable/plugins/
+```
+
+> The C code in `src/iop/sigmoid.c` uses `dlopen("libsigmoid_mojo.so", RTLD_LAZY | RTLD_LOCAL)` in `init_global` to load the Mojo module. 
diff --git a/src/external/OpenCL b/src/external/OpenCL
deleted file mode 160000
index 8a97ebc88d..0000000000
--- a/src/external/OpenCL
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit 8a97ebc88daa3495d6f57ec10bb515224400186f
diff --git a/src/external/OpenCL/.gitignore b/src/external/OpenCL/.gitignore
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/src/external/lua-scripts b/src/external/lua-scripts
deleted file mode 160000
index db505f1e0a..0000000000
--- a/src/external/lua-scripts
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit db505f1e0a089b210f5afa653e53ebd32e5a82b0
diff --git a/src/external/lua-scripts/.gitignore b/src/external/lua-scripts/.gitignore
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/src/iop/sigmoid.c b/src/iop/sigmoid.c
index cbd96d7725..b9301431b7 100644
--- a/src/iop/sigmoid.c
+++ b/src/iop/sigmoid.c
@@ -27,9 +27,11 @@
 #include "gui/gtk.h"
 #include "gui/presets.h"
 #include "iop/iop_api.h"
+#include "common/opencl.h"
 
 #include <gtk/gtk.h>
 #include <stdlib.h>
+#include <dlfcn.h>
 
 DT_MODULE_INTROSPECTION(3, dt_iop_sigmoid_params_t)
 
@@ -180,10 +182,38 @@ typedef struct dt_iop_sigmoid_gui_data_t
   dt_gui_collapsible_section_t display_luminance_section, primaries_section;
 } dt_iop_sigmoid_gui_data_t;
 
+typedef struct {
+  float white_target[4];
+  float black_target[4];
+  float paper_exposure[4];
+  float film_fog[4];
+  float film_power[4];
+  float paper_power[4];
+  float contrast_power[4];
+  float skew_power[4];
+  float hue_preservation[4];
+  float pipe_to_base[16];
+  float base_to_rendering[16];
+  float rendering_to_pipe[16];
+} SigmoidMojoParams;
+
+typedef void (*mojo_init_fn)(uintptr_t *ctx, int use_gpu);
+typedef void (*mojo_destroy_fn)(uintptr_t ctx);
+typedef void (*mojo_rgb_ratio_fn)(uintptr_t ctx, float *in, float *out, int32_t width, int32_t height, void *params);
+typedef void (*mojo_per_channel_fn)(uintptr_t ctx, float *in, float *out, int32_t width, int32_t height, void *params);
+
 typedef struct dt_iop_sigmoid_global_data_t
 {
   int kernel_sigmoid_loglogistic_per_channel;
   int kernel_sigmoid_loglogistic_rgb_ratio;
+  
+  void *mojo_lib;
+  uintptr_t mojo_ctx_cpu;
+  uintptr_t mojo_ctx_gpu;
+  mojo_init_fn mojo_init;
+  mojo_destroy_fn mojo_destroy;
+  mojo_rgb_ratio_fn mojo_rgb_ratio;
+  mojo_per_channel_fn mojo_per_channel;
 } dt_iop_sigmoid_global_data_t;
 
 
@@ -760,6 +790,40 @@ void process_loglogistic_per_channel(dt_develop_t *dev,
   }
 }
 
+static SigmoidMojoParams _build_mojo_params(dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece)
+{
+  const dt_iop_sigmoid_data_t *d = piece->data;
+  SigmoidMojoParams p = {0};
+
+  for(int i = 0; i < 4; i++) {
+    p.white_target[i] = d->white_target;
+    p.black_target[i] = d->black_target;
+    p.paper_exposure[i] = d->paper_exposure;
+    p.film_fog[i] = d->film_fog;
+    p.film_power[i] = d->film_power;
+    p.paper_power[i] = d->paper_power;
+    p.contrast_power[i] = d->film_power;
+    p.skew_power[i] = d->paper_power;
+    p.hue_preservation[i] = d->hue_preservation;
+  }
+
+  const dt_iop_order_iccprofile_info_t *pipe_work_profile = dt_ioppr_get_pipe_work_profile_info(piece->pipe);
+  const dt_iop_order_iccprofile_info_t *base_profile = _get_base_profile(self->dev, pipe_work_profile, d->base_primaries);
+  dt_colormatrix_t pipe_to_base_transposed, base_to_rendering_transposed, rendering_to_pipe_transposed;
+  dt_colormatrix_t pipe_to_base, base_to_rendering, rendering_to_pipe;
+  
+  _calculate_adjusted_primaries(d, pipe_work_profile, base_profile, pipe_to_base_transposed, base_to_rendering_transposed, rendering_to_pipe_transposed);
+  transpose_3xSSE(pipe_to_base_transposed, pipe_to_base);
+  transpose_3xSSE(base_to_rendering_transposed, base_to_rendering);
+  transpose_3xSSE(rendering_to_pipe_transposed, rendering_to_pipe);
+
+  memcpy(p.pipe_to_base, pipe_to_base, sizeof(pipe_to_base));
+  memcpy(p.base_to_rendering, base_to_rendering, sizeof(base_to_rendering));
+  memcpy(p.rendering_to_pipe, rendering_to_pipe, sizeof(rendering_to_pipe));
+
+  return p;
+}
+
 /** process, all real work is done here. */
 void process(dt_iop_module_t *self,
              dt_dev_pixelpipe_iop_t *piece,
@@ -768,8 +832,22 @@ void process(dt_iop_module_t *self,
              const dt_iop_roi_t *const roi_in,
              const dt_iop_roi_t *const roi_out)
 {
-  // this is called for preview and full pipe separately, each with its own pixelpipe piece.
   const dt_iop_sigmoid_data_t *module_data = piece->data;
+  const dt_iop_sigmoid_global_data_t *gd = self->global_data;
+
+  if(gd && gd->mojo_lib && gd->mojo_per_channel && gd->mojo_rgb_ratio)
+  {
+    SigmoidMojoParams p = _build_mojo_params(self, piece);
+    const int w = roi_in->width, h = roi_in->height;
+    int use_gpu = dt_opencl_is_enabled() ? 1 : 0;
+    uintptr_t ctx = use_gpu ? gd->mojo_ctx_gpu : gd->mojo_ctx_cpu;
+    
+    if(module_data->color_processing == DT_SIGMOID_METHOD_PER_CHANNEL)
+      gd->mojo_per_channel(ctx, (float *)ivoid, (float *)ovoid, w, h, &p);
+    else
+      gd->mojo_rgb_ratio(ctx, (float *)ivoid, (float *)ovoid, w, h, &p);
+    return;
+  }
 
   if(module_data->color_processing == DT_SIGMOID_METHOD_PER_CHANNEL)
   {
@@ -781,87 +859,42 @@ void process(dt_iop_module_t *self,
   }
 }
 
-#ifdef HAVE_OPENCL
-int process_cl(dt_iop_module_t *self,
-               dt_dev_pixelpipe_iop_t *piece,
-               cl_mem dev_in,
-               cl_mem dev_out,
-               const dt_iop_roi_t *const roi_in,
-               const dt_iop_roi_t *const roi_out)
-{
-  const dt_iop_sigmoid_data_t *const d = piece->data;
-  const dt_iop_sigmoid_global_data_t *const gd = self->global_data;
 
-  cl_int err = CL_MEM_OBJECT_ALLOCATION_FAILURE;
-  const int devid = piece->pipe->devid;
-  const int width = roi_in->width;
-  const int height = roi_in->height;
 
-  const float white_target = d->white_target;
-  const float paper_exp = d->paper_exposure;
-  const float film_fog = d->film_fog;
-  const float contrast_power = d->film_power;
-  const float skew_power = d->paper_power;
+void init_global(dt_iop_module_so_t *self)
+{
+  dt_iop_sigmoid_global_data_t *gd = calloc(1, sizeof(dt_iop_sigmoid_global_data_t));
+  self->data = gd;
 
-  const dt_iop_order_iccprofile_info_t *pipe_work_profile = dt_ioppr_get_pipe_work_profile_info(piece->pipe);
-  const dt_iop_order_iccprofile_info_t *base_profile = _get_base_profile(self->dev, pipe_work_profile, d->base_primaries);
-  dt_colormatrix_t pipe_to_base_transposed, base_to_rendering_transposed,
-      rendering_to_pipe_transposed, pipe_to_base, base_to_rendering, rendering_to_pipe;
-  _calculate_adjusted_primaries(d, pipe_work_profile, base_profile, pipe_to_base_transposed, base_to_rendering_transposed, rendering_to_pipe_transposed);
-  transpose_3xSSE(pipe_to_base_transposed, pipe_to_base);
-  transpose_3xSSE(base_to_rendering_transposed, base_to_rendering);
-  transpose_3xSSE(rendering_to_pipe_transposed, rendering_to_pipe);
-  const cl_mem dev_pipe_to_base
-      = dt_opencl_copy_host_to_device_constant(devid, sizeof(pipe_to_base), pipe_to_base);
-  const cl_mem dev_base_to_rendering
-      = dt_opencl_copy_host_to_device_constant(devid, sizeof(base_to_rendering), base_to_rendering);
-  const cl_mem dev_rendering_to_pipe
-      = dt_opencl_copy_host_to_device_constant(devid, sizeof(rendering_to_pipe), rendering_to_pipe);
-  if(dev_pipe_to_base == NULL || dev_base_to_rendering == NULL || dev_rendering_to_pipe == NULL)
-    goto cleanup;
-
-  if(d->color_processing == DT_SIGMOID_METHOD_PER_CHANNEL)
+  gd->mojo_lib = dlopen("libsigmoid_mojo.so", RTLD_LAZY | RTLD_LOCAL);
+  if(gd->mojo_lib)
   {
-    const float hue_preservation = d->hue_preservation;
-    err = dt_opencl_enqueue_kernel_2d_args(
-        devid, gd->kernel_sigmoid_loglogistic_per_channel, width, height, CLARG(dev_in), CLARG(dev_out),
-        CLARG(width), CLARG(height), CLARG(white_target), CLARG(paper_exp), CLARG(film_fog), CLARG(contrast_power),
-        CLARG(skew_power), CLARG(hue_preservation), CLARG(dev_pipe_to_base), CLARG(dev_base_to_rendering), CLARG(dev_rendering_to_pipe));
-  }
-  else
-  {
-    const float black_target = d->black_target;
+    gd->mojo_init = (mojo_init_fn)dlsym(gd->mojo_lib, "sigmoid_mojo_init");
+    gd->mojo_destroy = (mojo_destroy_fn)dlsym(gd->mojo_lib, "sigmoid_mojo_destroy");
+    gd->mojo_rgb_ratio = (mojo_rgb_ratio_fn)dlsym(gd->mojo_lib, "sigmoid_mojo_rgb_ratio");
+    gd->mojo_per_channel = (mojo_per_channel_fn)dlsym(gd->mojo_lib, "sigmoid_mojo_per_channel");
 
-    err = dt_opencl_enqueue_kernel_2d_args(devid, gd->kernel_sigmoid_loglogistic_rgb_ratio, width, height,
-                                           CLARG(dev_in), CLARG(dev_out), CLARG(width), CLARG(height),
-                                           CLARG(white_target), CLARG(black_target), CLARG(paper_exp),
-                                           CLARG(film_fog), CLARG(contrast_power), CLARG(skew_power));
+    if (gd->mojo_init && gd->mojo_destroy)
+    {
+      gd->mojo_init(&gd->mojo_ctx_cpu, 0); // CPU context
+      gd->mojo_init(&gd->mojo_ctx_gpu, 1); // GPU context
+    }
   }
-
-cleanup:
-  dt_opencl_release_mem_object(dev_pipe_to_base);
-  dt_opencl_release_mem_object(dev_base_to_rendering);
-  dt_opencl_release_mem_object(dev_rendering_to_pipe);
-  return err;
-}
-#endif // HAVE_OPENCL
-
-void init_global(dt_iop_module_so_t *self)
-{
-  const int program = 36; // sigmoid.cl, from programs.conf
-  dt_iop_sigmoid_global_data_t *gd = malloc(sizeof(dt_iop_sigmoid_global_data_t));
-
-  self->data = gd;
-  gd->kernel_sigmoid_loglogistic_per_channel = dt_opencl_create_kernel(program, "sigmoid_loglogistic_per_channel");
-  gd->kernel_sigmoid_loglogistic_rgb_ratio = dt_opencl_create_kernel(program, "sigmoid_loglogistic_rgb_ratio");
 }
 
 void cleanup_global(dt_iop_module_so_t *self)
 {
-  const dt_iop_sigmoid_global_data_t *gd = self->data;
-  dt_opencl_free_kernel(gd->kernel_sigmoid_loglogistic_per_channel);
-  dt_opencl_free_kernel(gd->kernel_sigmoid_loglogistic_rgb_ratio);
-  free(self->data);
+  dt_iop_sigmoid_global_data_t *gd = self->data;
+  if(gd)
+  {
+    if(gd->mojo_destroy)
+    {
+      if(gd->mojo_ctx_cpu) gd->mojo_destroy(gd->mojo_ctx_cpu);
+      if(gd->mojo_ctx_gpu) gd->mojo_destroy(gd->mojo_ctx_gpu);
+    }
+    if(gd->mojo_lib) dlclose(gd->mojo_lib);
+    free(gd);
+  }
   self->data = NULL;
 }