From 4a2f120668e9bbd65254ede795667ad3a39ac292 Mon Sep 17 00:00:00 2001 From: pi-os3 Date: Tue, 10 Nov 2026 19:31:15 -0400 Subject: [PATCH 1/2] docs/arch-gaps: four FLM families with no row, and a routing target that is closed MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two corrections on one page, both measured rather than judged. The 2026-11-08 and 2026-11-09 sections route the `!! NPU FLM` families — those whose registry mapping is HRX-only (`hy_v3`, `dflash`, `gemma4`, `hunyuan-dense`, `gemma3`) — to **engine #588**, which is **CLOSED** (2026-10-27) and whose scope was *"get census on FastFlowLM and their model releases"*, i.e. **watching**, not serving a family on the NPU route. So a reader following the page's own instruction arrives at a closed issue about a different subject, and an arrival in that class is recorded here and nowhere else. Both references now say so. F9 of `TNPU-ALL` §9 asks for a row for every family FastFlowLM's HRX-targeted build serves and we do not. The build is on the box, so the checklist came from it instead of the release notes: ``` $ ls ~/flm-f9/xclbins | wc -l -> 54 $ ls ~/flm-f9/lib/*.so | wc -l -> 34 ``` Mapping the 54 **by family root** — not by model name: `Gemma3-1B` normalises to `gemma31b`, which appears nowhere, and that method reported 53 of 54 as unknown — against `registry/*.json`, `docs/registry.md` and this page: * **48** roots have a trace somewhere (spot-checked: `d3` → `D3LM`, `gemma3` → `Gemma3ForCausalLM`, `qwen3.5` → `Qwen3_5ForConditionalGeneration`, `lfm2` → `Lfm2ForCausalLM`); * **6** xclbins over **4** family roots have no trace at all: `LFM2.5-1.2B` and `LFM2.5-1.2B-Thinking` (`lfm2.5`), `Medgemma-1.5-4B` and `Medgemma-4B` (`medgemma`), `Nanbeige4.1-3B` (`nanbeige4.1`), `Translategemma-4B-Instruct` (`translategemma`). `LFM2.5` is the sharpest: we map `Lfm2ForCausalLM` and `registry/significant.json` already carries *"lfm2-shaped config, different tensors"*, so a 2.5 generation is exactly the arrival this page exists to catch. It does not describe what any of the four families computes. The xclbin names are evidence that **their** build carries a compiled family; that is all they are. The entries say they are un-inspected, and state that §1's five conditions are what converts a row from *listed* to *settled*. Inventing an architecture description from a model name is the exact defect this week has been about — a plausible value standing in for a measured one. * `python3 tools/copyright.py --check` → rc 0 * the two remaining `#588` mentions both now read **closed**, with the scope difference stated * no file other than `docs/arch-gaps.md` is touched (`+43/−3`) Refs engine#1787 (F9, item 2 of its definition of done). --- docs/arch-gaps.md | 46 +++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 43 insertions(+), 3 deletions(-) diff --git a/docs/arch-gaps.md b/docs/arch-gaps.md index aa1d2cb0a..dc5d0d63e 100644 --- a/docs/arch-gaps.md +++ b/docs/arch-gaps.md @@ -1041,7 +1041,10 @@ takes the reviewed set: such a pack prints `. FLM ... registry: none (reviewed, Still open, and alerting on purpose: `MiniCPMV4_7ForConditionalGeneration` (needs one converter registration, not an architecture), `MiniCPMOcrForConditionalGeneration` (own `minicpm-ocr` library), `smolvla-NPU2` (no config at all), and the `!! NPU FLM` families whose registry mapping is HRX-only (`hy_v3`, `dflash`, `gemma4`, `hunyuan-dense`, `gemma3`) — -that last group is engine #588 (the family must be served on the NPU route), not a registry edit. +that last group is **not a registry edit**: the family has to be served on the NPU route. It used to be recorded here as +engine #588, which is **closed** (2026-10-27) and was about *watching FLM's releases*, not about serving a family on the +route — so this class currently has **no owning issue**, and an arrival that lands in it is recorded on this page and +nowhere else. See the 2026-11-10 section below for the arrivals that are sitting in it. ## Census run 2026-11-09 (`engine #1667`) @@ -1072,9 +1075,11 @@ Left as they are, on purpose, under the policy above: the six `. REMOTE-CODE` cl (`HajeriOperatorForCausalLM`, `MM200ForCausalLM`, `K2HorizonForCausalLM`, `KLMForCausalLM`, `KimiK31SuperExpertForCausalLM`, `ThalamusReasoningQwen3_5ForConditionalGeneration`); the `? UNVERIFIABLE` gated repos, which are retried on every run; and the `!! FLM` lines. Those keep the 2026-11-08 reasons: `MiniCPMV4_7`, `MiniCPMOcr` -and `smolvla` as above, and `!! NPU FLM` is engine #588. This run's issue window also includes the `gpt-oss` +and `smolvla` as above, and `!! NPU FLM` was recorded as engine #588 — which is **closed** (2026-10-27) and was +about watching FLM's releases, not about serving a family on the NPU route, so that class now has no owning issue +(corrected 2026-11-10; see the section at the end of this page). This run's issue window also includes the `gpt-oss` (`GPT-OSS-20B`, `GPT-OSS-Safeguard-20b`) and `lfm2` (`LFM2-1.2B`, `LFM2-2.6B`) Q4NX packs. Their mapping is HRX-only -too, so they belong to #588, not to a registry edit. +too, so they are the same class — a route to serve, not a registry edit. ## 2026-11-10: `DeepseekV4FlashSuperExpertForCausalLM` — reviewed, and recorded here @@ -1087,3 +1092,38 @@ the single undocumented entry out of 37. MLA-style geometry (`kv_lora_rank` 512, `q_lora_rank` 1536, `qk_rope_head_dim` 64, `v_head_dim` 128, 32 heads / 16 KV, IM 4096). **Not an alias of `deepseek_v41`**: nothing at the pin decodes AW-SECD weights, so mapping it to that architecture would name a backend that cannot read the file. Reported, not alerted (`"watch": "report"`). + +## FLM 1.0.12's HRX build: four family roots with no row (engine#1787, 2026-11-10) + +F9 of `TNPU-ALL` §9 asks for a row for *every family FastFlowLM's HRX-targeted build serves and we do not*. The build is +on the box, so the checklist was enumerated from it rather than from the release notes: + +``` +$ ls ~/flm-f9/xclbins | wc -l -> 54 (per-family xclbin directories) +$ ls ~/flm-f9/lib/*.so | wc -l -> 34 (per-family shared objects) +``` + +Mapping each of the 54 by **family root** (not by model name: `Gemma3-1B` normalises to `gemma31b`, which appears nowhere, +and matching that way reported 53 of 54 as unknown) against `registry/*.json`, `docs/registry.md` and this page: + +* **48** roots have a trace somewhere — spot-checked rather than assumed: `d3` → `D3LM` in `census.json`, `gemma3` → + `Gemma3ForCausalLM` in `architectures.json`, `qwen3.5` → `Qwen3_5ForConditionalGeneration`, `lfm2` → `Lfm2ForCausalLM`. +* **6** xclbins over **4** family roots have **no trace at all**: + +| xclbin (FLM 1.0.12 HRX) | family root | state | +|---|---|---| +| `LFM2.5-1.2B` | `lfm2.5` | no row, no registry entry. We map `Lfm2ForCausalLM`, and `registry/significant.json` already carries the note *"lfm2-shaped config, different tensors"*, so a 2.5 generation is exactly the arrival this page exists to catch | +| `LFM2.5-1.2B-Thinking` | `lfm2.5` | as above | +| `Medgemma-1.5-4B` | `medgemma` | no row, no registry entry | +| `Medgemma-4B` | `medgemma` | as above | +| `Nanbeige4.1-3B` | `nanbeige4.1` | no row, no registry entry | +| `Translategemma-4B-Instruct` | `translategemma` | no row, no registry entry | + +**These rows are un-inspected, and say so on purpose.** Nothing above describes what any of the four families computes: the +xclbin names are evidence that *their* build carries a compiled family for it, and that is all. §1's five conditions are +what converts an entry from *listed* to *settled* — imports with zero gaps and serving on the device, a known-answer prompt +under the model's own template with three identical answers, decode tok/s and TTFT under a ticket, and the kernel set and +engine commit named. Writing an architecture description from a model name is the defect this week has been about; a row +that is honest about being un-inspected is more useful than a plausible paragraph. + +The `!! NPU FLM` class above is where these belong once someone owns it. As of this run that class has no owning issue. From 5c1d8e85fafa23ede673b0fe77b95da510903ae4 Mon Sep 17 00:00:00 2001 From: bong-water-water-bong Date: Wed, 11 Nov 2026 11:02:47 -0400 Subject: [PATCH 2/2] =?UTF-8?q?docs/arch-gaps:=20the=20F9=20family=20list?= =?UTF-8?q?=20restated=20=E2=80=94=20one=20root=20with=20nothing,=20three?= =?UTF-8?q?=20already=20mapped=20or=20served?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Answers the review of #2065: the 'no trace' table compared family roots as literals against class names. nanbeige4.1 is mapped, medgemma is a gemma3 arrival with a row, lfm2.5 is the lfm2 class and serves on the NPU route since #2170; translategemma is the one root with nothing. The class's owning issue is #1786, and the xclbins are NPU artifacts. Co-Authored-By: Claude Fable 5.2 --- docs/arch-gaps.md | 67 ++++++++++++++++++++++++++--------------------- 1 file changed, 37 insertions(+), 30 deletions(-) diff --git a/docs/arch-gaps.md b/docs/arch-gaps.md index dc5d0d63e..33f8a1a84 100644 --- a/docs/arch-gaps.md +++ b/docs/arch-gaps.md @@ -1043,8 +1043,8 @@ architecture), `MiniCPMOcrForConditionalGeneration` (own `minicpm-ocr` library), the `!! NPU FLM` families whose registry mapping is HRX-only (`hy_v3`, `dflash`, `gemma4`, `hunyuan-dense`, `gemma3`) — that last group is **not a registry edit**: the family has to be served on the NPU route. It used to be recorded here as engine #588, which is **closed** (2026-10-27) and was about *watching FLM's releases*, not about serving a family on the -route — so this class currently has **no owning issue**, and an arrival that lands in it is recorded on this page and -nowhere else. See the 2026-11-10 section below for the arrivals that are sitting in it. +route. The issue that does own this class is engine#1786 (every `FastFlowLM/*-NPU2` source serves on `--device npu`, +opened 2026-11-10). See the last section of this page for the arrivals that are sitting in it. ## Census run 2026-11-09 (`engine #1667`) @@ -1076,8 +1076,8 @@ Left as they are, on purpose, under the policy above: the six `. REMOTE-CODE` cl `KimiK31SuperExpertForCausalLM`, `ThalamusReasoningQwen3_5ForConditionalGeneration`); the `? UNVERIFIABLE` gated repos, which are retried on every run; and the `!! FLM` lines. Those keep the 2026-11-08 reasons: `MiniCPMV4_7`, `MiniCPMOcr` and `smolvla` as above, and `!! NPU FLM` was recorded as engine #588 — which is **closed** (2026-10-27) and was -about watching FLM's releases, not about serving a family on the NPU route, so that class now has no owning issue -(corrected 2026-11-10; see the section at the end of this page). This run's issue window also includes the `gpt-oss` +about watching FLM's releases, not about serving a family on the NPU route; the class is owned by engine#1786 +(corrected 2026-11-11; see the section at the end of this page). This run's issue window also includes the `gpt-oss` (`GPT-OSS-20B`, `GPT-OSS-Safeguard-20b`) and `lfm2` (`LFM2-1.2B`, `LFM2-2.6B`) Q4NX packs. Their mapping is HRX-only too, so they are the same class — a route to serve, not a registry edit. @@ -1093,37 +1093,44 @@ the single undocumented entry out of 37. 16 KV, IM 4096). **Not an alias of `deepseek_v41`**: nothing at the pin decodes AW-SECD weights, so mapping it to that architecture would name a backend that cannot read the file. Reported, not alerted (`"watch": "report"`). -## FLM 1.0.12's HRX build: four family roots with no row (engine#1787, 2026-11-10) +## FLM 1.0.12's build on the box: one family root with nothing, three that are already somewhere (engine#1787) -F9 of `TNPU-ALL` §9 asks for a row for *every family FastFlowLM's HRX-targeted build serves and we do not*. The build is -on the box, so the checklist was enumerated from it rather than from the release notes: +F9 of `TNPU-ALL` §9 asks for a row for *every family FastFlowLM's build serves and we do not*. The build is on the box, +so the checklist was enumerated from it rather than from the release notes: ``` $ ls ~/flm-f9/xclbins | wc -l -> 54 (per-family xclbin directories) $ ls ~/flm-f9/lib/*.so | wc -l -> 34 (per-family shared objects) ``` -Mapping each of the 54 by **family root** (not by model name: `Gemma3-1B` normalises to `gemma31b`, which appears nowhere, -and matching that way reported 53 of 54 as unknown) against `registry/*.json`, `docs/registry.md` and this page: - -* **48** roots have a trace somewhere — spot-checked rather than assumed: `d3` → `D3LM` in `census.json`, `gemma3` → - `Gemma3ForCausalLM` in `architectures.json`, `qwen3.5` → `Qwen3_5ForConditionalGeneration`, `lfm2` → `Lfm2ForCausalLM`. -* **6** xclbins over **4** family roots have **no trace at all**: +A word on the name: #1787 calls this "FLM 1.0.12's HRX build", and the section used to. What is counted here is a +directory of **xclbins, every one named `*-NPU2`**: they are NPU artifacts. A reader looking for GPU families will not +find them in this list. -| xclbin (FLM 1.0.12 HRX) | family root | state | -|---|---|---| -| `LFM2.5-1.2B` | `lfm2.5` | no row, no registry entry. We map `Lfm2ForCausalLM`, and `registry/significant.json` already carries the note *"lfm2-shaped config, different tensors"*, so a 2.5 generation is exactly the arrival this page exists to catch | -| `LFM2.5-1.2B-Thinking` | `lfm2.5` | as above | -| `Medgemma-1.5-4B` | `medgemma` | no row, no registry entry | -| `Medgemma-4B` | `medgemma` | as above | -| `Nanbeige4.1-3B` | `nanbeige4.1` | no row, no registry entry | -| `Translategemma-4B-Instruct` | `translategemma` | no row, no registry entry | - -**These rows are un-inspected, and say so on purpose.** Nothing above describes what any of the four families computes: the -xclbin names are evidence that *their* build carries a compiled family for it, and that is all. §1's five conditions are -what converts an entry from *listed* to *settled* — imports with zero gaps and serving on the device, a known-answer prompt -under the model's own template with three identical answers, decode tok/s and TTFT under a ticket, and the kernel set and -engine commit named. Writing an architecture description from a model name is the defect this week has been about; a row -that is honest about being un-inspected is more useful than a plausible paragraph. - -The `!! NPU FLM` class above is where these belong once someone owns it. As of this run that class has no owning issue. +Mapping each of the 54 by **family root** (not by model name: `Gemma3-1B` normalises to `gemma31b`, which appears nowhere, +and matching that way reported 53 of 54 as unknown) against `registry/*.json`, `docs/registry.md` and this page gave 48 +with a trace and 6 xclbins over 4 roots without one. **That second number was wrong** (review of engine#2065): the root +was compared as a literal against class names and `model_type`s that do not contain it (`nanbeige4.1` against +`NanbeigeForCausalLM`, `medgemma` against `gemma3`). Each of the four, searched on `main` case-insensitively and +restated 2026-11-11: + +| xclbin directory | family root | what `main` has | what is actually owed | +|---|---|---|---| +| `Translategemma-4B-Instruct-NPU2` | `translategemma` | nothing: `git grep -il translategemma` finds 0 files | everything. This is the one root with no trace. By its name it would be a `gemma3` arrival like `medgemma`; nobody has opened its config, so that is a guess and not a mapping | +| `LFM2.5-1.2B-NPU2`, `LFM2.5-1.2B-Thinking-NPU2` | `lfm2.5` | the packs' own `config.json` says `Lfm2ForCausalLM` / `model_type: lfm2`, which the registry maps, so there is no "2.5 class" to add; `models/pins.json` pins the `LFM2.5-2.6B` and `LFM2.5-8B-A1B` GGUFs; and since engine#2170 both packs **serve on the NPU route** with measured rows in [npu.md](npu.md) ("FastFlowLM's own containers") | nothing in the registry. On the route: speed (per-op, about 4 tok/s against their 48 to 83), not coverage | +| `Medgemma-4B-NPU2`, `Medgemma-1.5-4B-NPU2` | `medgemma` | a `gemma3` arrival, already routed: [npu.md](npu.md) has the row `gemma3` / `medgemma-1.5-4b-it-NPU2` / hrx / not served, and `tests/census_watch_test.py` pins that pack to `Gemma3ForCausalLM` | an NPU row. `Medgemma-4B-NPU2` is on the box; its geometry (H 2560, IM 10240) needs two kernel shapes that are not built (`GU K2560 N20480`, `D K10240 N2560`), so it is a kernel-set item, not a model-type one | +| `Nanbeige4.1-3B-NPU2` | `nanbeige4.1` | mapped: `registry/architectures.json` has `NanbeigeForCausalLM` (gguf `nanbeige`), `registry/census.json` counts it, `registry/support_pins.json` lists its converter config, and `q4nx_semantics.h` and [tokenizers.md](tokenizers.md) name `Nanbeige4.1-3B` itself (11 files) | an NPU row. Nothing on [npu.md](npu.md) says whether it serves there, and it has not been tried | + +So: **one** family root with nothing (`translategemma`), and three that are mapped or served and owe a measured NPU row +or nothing at all. The count of 48 "with a trace" was spot-checked, not re-derived, and after this correction it should +be read as "at least 51". + +**The rows that are still owed are un-inspected, and say so on purpose.** Nothing above describes what `translategemma` +or `nanbeige4.1` computes: an xclbin name is evidence that *their* build carries a compiled family for it, and that is +all. §1's five conditions are what converts an entry from *listed* to *settled* — imports with zero gaps and serving on +the device, a known-answer prompt under the model's own template with three identical answers, decode tok/s and TTFT +under a ticket, and the kernel set and engine commit named. The two `lfm2.5` packs are the example of an entry that +went from listed to settled that way. + +The `!! NPU FLM` class above is where the owed rows belong. Its owning issue is engine#1786 (every `*-NPU2` source +serves), which is open.