From ba9f9ee436355f9d74e58d21498b1085fe2bd61d Mon Sep 17 00:00:00 2001 From: pi-tmplsec Date: Tue, 10 Nov 2026 13:00:32 -0400 Subject: [PATCH] census: DeepseekV4FlashSuperExpertForCausalLM is reviewed and report-only (engine#1922) The pack's own config.json settles it two ways: `auto_map` points at its own modeling_ module (remote code, so there is no GGUF architecture for a pinned upstream to accept), and it ships AW-SECD-compressed weights (8.5 GB, claimed 0.962 retention), so nothing at the pin could decode the file even with the architecture mapped. Recorded in significant.json with watch: report, i.e. listed and not alerted. Pinned by four checks in tests/census_watch_test.py: report-only, the reason names the compression rather than just the decline, the class stays out of architectures.json (which is generated from the pin and must not gain a guessed gguf name), and the model_type is recorded. K2HorizonForCausalLM is deliberately NOT added: the repo is gated (401 on both the raw config and the model API), so there is no model_type to record and a reviewed entry would be a claim about a class nobody has read. --- registry/significant.json | 5 +++++ tests/census_watch_test.py | 12 ++++++++++++ 2 files changed, 17 insertions(+) diff --git a/registry/significant.json b/registry/significant.json index e9dc99d82..75ac3b192 100644 --- a/registry/significant.json +++ b/registry/significant.json @@ -167,6 +167,11 @@ "model_type": "nanogpt-custom", "family": "nanoGPT-style one-offs (first seen: learner-subodh/Delivery-Rider-Assistant-SLM). 'GPT' is the class name nanoGPT's model.py exports, so this key covers any upload that saves nanoGPT's own state dict. Not an alias of GPT2LMHeadModel: the config uses nanoGPT's keys (n_embd 384, n_layer 6, n_head 8, block_size 256, bias, dropout) rather than GPT2Config's n_positions/n_ctx, the repo carries only a pickled pytorch_model.bin (no safetensors, no tokenizer files; the card names tiktoken gpt2), and nanoGPT stores nn.Linear weights where GPT-2's checkpoint stores Conv1D (transposed), so the gpt2 converter would read every projection the wrong way round. Nothing at the pin registers the class. Declined on 2026-11-09 under engine #1667; the census reports it as . REVIEWED. Remove this entry if a backend gains a nanoGPT import.", "watch": "report" + }, + "DeepseekV4FlashSuperExpertForCausalLM": { + "model_type": "deepseek_v4_flash_superexpert", + "family": "DeepSeek V4.1 Flash SuperExpert, published as remote code and with AW-SECD-compressed weights (8.5 GB, claimed 0.962 retention); 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 would name a backend that cannot read it", + "watch": "report" } } } diff --git a/tests/census_watch_test.py b/tests/census_watch_test.py index e70f10803..04f04b906 100644 --- a/tests/census_watch_test.py +++ b/tests/census_watch_test.py @@ -937,6 +937,18 @@ def _fake_urlopen(req, timeout=0): check("a 401 falls back to anonymous instead of failing the run", page == [{"id": "x/y"}] and seen_auth[0] == "Bearer hf_stale" and seen_auth[1] is None) + +# 17. engine#1922: DeepSeek V4.1 Flash SuperExpert is reviewed and report-only -- listed, not alerted. +# The class is remote code (its own modeling_ module) AND ships AW-SECD-compressed weights, so nothing at +# the pin decodes it: mapping it in architectures.json would name a backend that cannot read the file. +_DS = "DeepseekV4FlashSuperExpertForCausalLM" +check("engine#1922 class is recorded as report-only", + sig.get(_DS, {}).get("watch") == "report") +check("engine#1922 entry states WHY (the compressed weights), not just that it is declined", + "AW-SECD" in sig.get(_DS, {}).get("family", "")) +check("engine#1922 class stays out of the runtime registry", _DS not in archs) +check("engine#1922 entry records its model_type", sig.get(_DS, {}).get("model_type") == "deepseek_v4_flash_superexpert") + if failures: print(f"\n{len(failures)} check(s) failed") sys.exit(1)