DeepSeek V3 FP8 Support - #2719

Merged

nv-guomingz merged 14 commits into

NVIDIA:deepseekfrom

yingcanw:deepseek

Jan 24, 2025

Commits on Dec 24, 2024

Add README
yingcanw
committed
ac56e4e
Add unified converter (#1)

jershi425
and
uchihatmtkinu
authored
0614867

Commits on Dec 25, 2024

update README
yingcanw
committed
da62d7b
add FP8 notes
yingcanw
committed
911c30e
Update run.py result
Hanyue He
committed
e3077bc

Commits on Dec 26, 2024

Update V3 README
yingcanw
committed
723eb27
Update usages of FP8 to BF16 instruction
yingcanw
committed
48844c4

Commits on Jan 2, 2025

fix model name mapping (#2)
jershi425
authored
9522bbc

Commits on Jan 6, 2025

Update HF ckpt BF16 conversion.
root
committed
281fd3a
Merge branch 'deepseek' into deepseek
yingcanw
authored
e0ce79f

Commits on Jan 16, 2025

fix config of deepseek kv cache
yingcanw
authored
0321f71

Commits on Jan 24, 2025

Remove source code
yingcanw
committed
ed789c6
Deepseek V3 FP8 Support
yingcanw
committed
8539574
Merge branch 'deepseek' into deepseek
yingcanw
authored
90972eb