Qwen3.5-122B-A10B on DGX Spark: 28.3 → 51 tok/s (+80%)

DGX Spark research and tests - containers, benchmarks, and investigation notes for running models on GB10 (SM 12.1)

7.67× LoRA / 8.35× Full FT speedup for Qwen3.5 (0.8B–27B) on NVIDIA DGX Spark — wall-clock parity with rented H100. Lossless within BF16. Three-command interactive wizard handles model picker, data validator, training, and merge.

Optimized CUDA kernels for NVIDIA GB10 Blackwell (sm_121, DGX Spark). RMSNorm + GELU. First sm_121 kernel on HuggingFace Kernel Hub.

DGX Spark (GB10/SM121) platform support for Meta's KernelAgent — auto-detect, hardware constraints, safe Triton configs

Empirical kernel scheduling characterization for NVIDIA GB10 (SM121a). Sweeps GEMM tile configurations, classifies PTX instruction paths, captures hardware telemetry

Pre-built PyTorch wheels and build scripts for NVIDIA DGX Spark (GB10, sm_121, Blackwell, CUDA 13.0, ARM64)

Patches + recipe to deploy festr2/MiMo-V2.5-Pro-NVFP4-MXFP8-attn-TP8 on 8-node DGX Spark sm_121 (Ray + vLLM, TP=8). Fixes the fused-qkv loader bug that mis-slotted Q values as K/V on 7 of 8 ranks.