Repository navigation
knowledge: Qwen3-30B-A3B quantized checkpoints on vLLM 0.30; H200 venue image differences - #99
Closed
AdvayMonga wants to merge 1 commit into
Closed
AdvayMonga wants to merge 1 commit into
AdvayMonga wants to merge 1 commit into
Conversation
…, loadability); H200 venue image differences Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Two findings from BlameGraph's correctness-gate runs (2026-10-04 Verda, 2026-10-05 Nebius). Data only, no code.
kb-20261005-ea8b253e: which quantized Qwen3-30B-A3B checkpoints keep >= 99% of the BF16 reference's pooled task score on 18,606 items (FP8 with and without DeepGEMM, INT8nytopop/Qwen3-30B-A3B.w8a8pass; GPTQ-Int4 fails at 0.9869, all of it MMLU-Pro), and that GPTQ-Int8 doesn't load on vLLM 0.30.kb-20261005-7785a33d: Verda vs Nebius H200 images (nvcc 12.8 vs 13.0, so DeepGEMM only JITs on Nebius; Python.h missing on Verda), plus the unresolved observation that plain BF16 matched the batch-invariant reference exactly on Nebius at concurrency 128 but diverged widely on Verda at concurrency 32.Evidence links to the run summary in BlameGraph's CLAUDE.md; the raw reports are local to BlameGraph's
data/. Both files seed cleanly (python -m lab.ledger seed), andtests/test_lab_ledger.pypasses.🤖 Generated with Claude Code