ci(benchmark): upgrade Kimi K2.5 to K2.6#781
Open
carlushuang wants to merge 1 commit into
Open
Conversation
Replace Kimi-K2.5-MXFP4 with Kimi-K2.6-MXFP4 across benchmark and accuracy configs. Also update Eagle3 draft model to kimi-k2.6-eagle3. Same architecture (KimiK25ForConditionalGeneration), better weights. Accuracy baseline from HF amd/Kimi-K2.6-MXFP4: gsm8k=0.9393 (BF16), MXFP4=0.9318 (99.2% recovery). Quantized by AMD via Quark. Note: K2.6-MXFP4 requires ATOM with PR#744 (MXFP4 shuffle alignment) and latest aiter with topk_gating.
carlushuang
force-pushed
the
carhuang/update_kimi_k2.6
branch
from
814a70a to
ee394bb
Compare
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Replace Kimi-K2.5-MXFP4 with Kimi-K2.6-MXFP4 across all CI configs.
Same architecture (
KimiK25ForConditionalGeneration,model_type: kimi_k25), better-trained weights. No model code changes needed.Changes
amd/Kimi-K2.6-MXFP4Accuracy baseline from HF card: gsm8k=0.9393 (BF16), MXFP4=0.9318 (99.2% recovery).
Files updated:
.github/benchmark/models.json.github/benchmark/models_accuracy.json.github/benchmark/oot_models_accuracy.jsonTest plan
KimiK25ForConditionalGenerationclass