vllm
[Feature][ROCm][Quantization] Requantize DeepSeek-V4.1 MXFP4 experts to block FP8 at load time
#69240
Open
Go
Login via GitHub
Home
FAQ
Install
Login
via GitHub
Overview
Commits
4
Changes
View On
GitHub
[Feature][ROCm][Quantization] Requantize DeepSeek-V4.1 MXFP4 experts to block FP8 at load time
#69240
yuzhouo7
wants to merge 4 commits into
vllm-project:main
from
yuzhouo7:rocm/dsv41-fp8-forward-parity
[Feature][ROCm][Quantization] Requantize MXFP4 MoE weights to block F…
c2411412
[ROCm][Quantization] Guard block FP8 requantization on clamped SiLU
02fcec37
[Spec Decode] Apply online quantization targets to embedded DeepSeek …
343e66d8
[ROCm] Match DeepSeek V4.1 TP8 online block-FP8 to checkpoint inference
3710f2be
mergify
added
documentation
mergify
added
deepseek
mergify
added
quantization
mergify
added
rocm
mergify
added
speculative-decoding
mergify
added
dflash
mergify
added
mrv2
yuzhouo7
marked this pull request as ready for review
3 hours ago
yuzhouo7
requested a review
from
tjtanaa
3 hours ago
yuzhouo7
requested a review
from
dllehr-amd
3 hours ago
yuzhouo7
requested a review
from
hongxiayang
3 hours ago
yuzhouo7
requested a review
from
AndreasKaratzas
3 hours ago
yuzhouo7
requested a review
from
mgoin
3 hours ago
yuzhouo7
requested a review
from
robertgshaw2-redhat
3 hours ago
yuzhouo7
requested a review
from
tlrmchlsmth
3 hours ago
yuzhouo7
requested a review
from
yewentao256
3 hours ago
claude
commented on 2026-11-22
yuzhouo7
requested a review
from
pavanimajety
3 hours ago
yuzhouo7
requested a review
from
zyongye
3 hours ago
yuzhouo7
requested a review
from
WoosukKwon
3 hours ago
yuzhouo7
requested a review
from
njhill
3 hours ago
yuzhouo7
requested a review
from
TheEpicDolphin
3 hours ago
yuzhouo7
requested a review
from
wzhao18
3 hours ago
yuzhouo7
requested a review
from
bnellnm
3 hours ago
Login to write a write a comment.
Login via GitHub
Reviewers
claude
tjtanaa
dllehr-amd
hongxiayang
AndreasKaratzas
mgoin
robertgshaw2-redhat
tlrmchlsmth
yewentao256
pavanimajety
zyongye
WoosukKwon
njhill
TheEpicDolphin
wzhao18
bnellnm
Assignees
No one assigned
Labels
documentation
rocm
speculative-decoding
deepseek
quantization
mrv2
dflash
Milestone
No milestone
Login to write a write a comment.
Login via GitHub