Add small-model GGUF runtime evidence - #607
Open
justinchuby wants to merge 1 commit into
Open
Conversation
Add pinned SmolLM and SmolLM2 F16 GGUF L4/L5 cases with immutable artifact, reference, tensor-census, import-route, and golden provenance checks. Exercise CLI save/reload, full-logit prefill and cache-threaded decode parity, deterministic generation, and wire the cases into affected-model and fast integration CI selection. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Signed-off-by: Justin Chu <justinchuby@users.noreply.github.com>
Performance Comparison
|
This was referenced Aug 25, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
HuggingFaceTB/SmolLM-135MandHuggingFaceTB/SmolLM2-135M-Instructthrough real CLI build, save/reload, ORT prefill, 20 cache-threaded decode steps, full-logit HF parity, and deterministic generationllamain affected-model detection and run the real-artifact tests in fast integration CIEvidence
neopolita/smollm-135m-gguf@22cca988936eafe92908e7558907c3964e10bba7/ggml-model-f16.ggufec8c775c16944a7e4b5251f97b3f848500dcc3e701b0d492ce9055cea42138a2F16=211,F32=61)unsloth/SmolLM2-135M-Instruct-GGUF@9e6855bc4be717fca1ef21360a1db4b29d5c559a/SmolLM2-135M-Instruct-F16.gguf5157ca60744d21631818364854ac8e4452e1b8022d2ab4c8a2f9cda2344afb30F16=211,F32=61)Both routes are CPU, dynamic-cache,
llama_qk_permute=true, andpreserve_quantization=false. Runtime packaging remains deferred: the public GGUFs use compiled llama.cppsmollmpre-tokenizer metadata and do not embedtokenizer.huggingface.json, so no architecture-wide or runtime-package support is claimed.The pinned SmolLM2 Q4_K_M artifact (
105,454,144bytes,ed5fa30c487b282ec156c29062f1222e5c20875a944ac98289dbd242e947f747) was exercised but is not promoted or compared against mismatched unquantized weights. Its preserved MatMulNBits route diverges from the same artifact under llama.cpp, while Mobius dequantization matches llama.cpp; this exact blocker is documented in the integration test.Validation
The broad non-integration suite remains at 7,054 passed with two pre-existing
glm_moe_dsashape-inference failures unrelated to this diff. CI was not awaited.