microsoft/DeepSpeed

Pull Requests Commits

Merge branch 'add-llama2-support' into quantization-refresh

RezaYazdaniAminabadi committed 1 year ago

Verified e2ef102f

Merge branch 'master' into add-llama2-support

RezaYazdaniAminabadi committed 1 year ago

Verified c72aa76a

Merge branch 'add-llama2-support' of github.com:microsoft/DeepSpeed into add-llama2-support

Reza Yazdani committed 1 year ago

10a1df25

fix an issue with the float32 transform kernel

Reza Yazdani committed 1 year ago

a87860d2

fix deepspeed4science links (#4358)

conglongli committed 1 year ago

Verified 78c3b148

fix syntax and format errors.

awan-10 committed 1 year ago

297a15cd

Merge branch 'master' into add-llama2-support

awan-10 committed 1 year ago

Verified d0abfdda

use the num_kv param only if it is positive

Reza Yazdani committed 1 year ago

db5a3b7f

DeepSpeed4Science (#4357)

conglongli committed 1 year ago

Verified f876d81d

Support InternLM (#4137)

wangruohui committed 1 year ago

Verified 367d6f9c

Merge branch 'add-llama2-support' of github.com:microsoft/DeepSpeed into add-llama2-support

Reza Yazdani committed 1 year ago

ca61bd1e

use num_kv only when it has positive value

Reza Yazdani committed 1 year ago

c33bc4fd

Merge branch 'master' into add-llama2-support

RezaYazdaniAminabadi committed 1 year ago

Verified b37f7d86

Merge branch 'add-llama2-support' into quantization-refresh

RezaYazdaniAminabadi committed 1 year ago

Verified 66e97f81

Fix DS-Inference quantization and add more bits support

Reza Yazdani committed 1 year ago

9f71c2a1

Pass base_dir to model files can be loaded for auto-tp/meta-tensor. (#4348)

awan-10 committed 1 year ago

Verified b9d719a6

Merge branch 'master' into add-llama2-support

mrwyattii committed 1 year ago

Verified 3e3945eb

bring back quantization and add different bits support

Reza Yazdani committed 1 year ago

f611c670

Fix autotune to support Triton 2.1 (#4340)

stephen-youn committed 1 year ago

Verified 3f3e9fb1

Merge branch 'master' into add-llama2-support

RezaYazdaniAminabadi committed 1 year ago

Verified c0ca80e8

Fix skipped inference tests (#4336)

mrwyattii committed 1 year ago

Verified 8c7f7fd2

Suppress noise (#4310)

tjruwase committed 1 year ago

Verified 3ea42c6e

Fix a bug in the implementation of dequantization for inference (#3433)

sakogan committed 1 year ago

Verified 9bf77782

DS-Chat BLOOM: Fix Attention mask (#4338)

lekurile committed 1 year ago

Verified 0c83436a

clear redundant timers (#4308)

starkhu committed 1 year ago

Verified e20e4a9d

Add release version checking (#4328)

loadams committed 1 year ago

Verified b5977d41

mrwyattii committed 1 year ago

24a3a0f2

Fix Zero3 contiguous grads, reduce scatter false accuracy issue (#4321)

nelyahu committed 1 year ago

Verified 11aa880e

Merge branch 'master' into add-llama2-support

RezaYazdaniAminabadi committed 1 year ago

Verified 81d692d0

remove more debugging code

Reza Yazdani committed 1 year ago

165042df

Older