vllm
[Feature][ROCm][Quantization] Requantize DeepSeek-V4.1 MXFP4 experts to block FP8 at load time
#69240
Open

[Feature][ROCm][Quantization] Requantize DeepSeek-V4.1 MXFP4 experts to block FP8 at load time #69240

yuzhouo7
yuzhouo7 [Feature][ROCm][Quantization] Requantize MXFP4 MoE weights to block F…
c2411412
yuzhouo7 [ROCm][Quantization] Guard block FP8 requantization on clamped SiLU
02fcec37
yuzhouo7 [Spec Decode] Apply online quantization targets to embedded DeepSeek …
343e66d8
yuzhouo7 [ROCm] Match DeepSeek V4.1 TP8 online block-FP8 to checkpoint inference
3710f2be
mergify
mergify mergify added documentation
mergify mergify added deepseek
mergify mergify added quantization
mergify mergify added rocm
mergify mergify added speculative-decoding
mergify mergify added dflash
mergify mergify added mrv2
yuzhouo7 yuzhouo7 marked this pull request as ready for review 3 hours ago
yuzhouo7 yuzhouo7 requested a review from tjtanaa tjtanaa 3 hours ago
yuzhouo7 yuzhouo7 requested a review from dllehr-amd dllehr-amd 3 hours ago
yuzhouo7 yuzhouo7 requested a review from hongxiayang hongxiayang 3 hours ago
yuzhouo7 yuzhouo7 requested a review from AndreasKaratzas AndreasKaratzas 3 hours ago
yuzhouo7 yuzhouo7 requested a review from mgoin mgoin 3 hours ago
yuzhouo7 yuzhouo7 requested a review from robertgshaw2-redhat robertgshaw2-redhat 3 hours ago
yuzhouo7 yuzhouo7 requested a review from tlrmchlsmth tlrmchlsmth 3 hours ago
yuzhouo7 yuzhouo7 requested a review from yewentao256 yewentao256 3 hours ago
claude
claude commented on 2026-11-22
yuzhouo7 yuzhouo7 requested a review from pavanimajety pavanimajety 3 hours ago
yuzhouo7 yuzhouo7 requested a review from zyongye zyongye 3 hours ago
yuzhouo7 yuzhouo7 requested a review from WoosukKwon WoosukKwon 3 hours ago
yuzhouo7 yuzhouo7 requested a review from njhill njhill 3 hours ago
yuzhouo7 yuzhouo7 requested a review from TheEpicDolphin TheEpicDolphin 3 hours ago
yuzhouo7 yuzhouo7 requested a review from wzhao18 wzhao18 3 hours ago
yuzhouo7 yuzhouo7 requested a review from bnellnm bnellnm 3 hours ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone