Commits

Commits on Jul 17, 2023

Merge pull request #4 from yandai/llama
Show description for e770ddf
void-main
authored
e770ddf
Copy full SHA for e770ddf

Commits on Jul 16, 2023

remove padded_embedding_bias_
yandai
committed
f1dd8fb
Copy full SHA for f1dd8fb

Commits on Jul 8, 2023

fix llama
yandai
committed
99a844c
Copy full SHA for 99a844c

Commits on Jul 7, 2023

Merge pull request #3 from CN-COTER/dev_support_int8_share_ctx
Show description for 3469ff0
void-main
authored
3469ff0
Copy full SHA for 3469ff0
fix bug: ft-llama-int8 output is incorrect
shaoxin
committed
c137bb5
Copy full SHA for c137bb5

Commits on Jul 3, 2023

Merge pull request #2 from CN-COTER/dev_support_int8_share_ctx
Show description for 69de159
void-main
authored
69de159
Copy full SHA for 69de159
upload code: llama int8&share_context triton backend
shaoxin
committed
9e374ef
Copy full SHA for 9e374ef

Commits on Jul 2, 2023

Merge pull request #1 from CN-COTER/dev_support_int8_share_ctx
Show description for 98435f1
void-main
authored
98435f1
Copy full SHA for 98435f1

Commits on Jun 30, 2023

rm unused code
shaoxin
committed
0e6ae5b
Copy full SHA for 0e6ae5b
update code with comments from reviewer
shaoxin
committed
6f0469f
Copy full SHA for 6f0469f

Commits on Jun 29, 2023

support int8 & share context
shaoxin
committed
3c03564
Copy full SHA for 3c03564

Commits on Jun 11, 2023

fix name
void-main
committed
50fdb0c
Copy full SHA for 50fdb0c

Commits on May 6, 2023

Merge branch 'NVIDIA:main' into main
void-main
authored
b942806
Copy full SHA for b942806
fix callback issue
void-main
committed
abd1e4d
Copy full SHA for abd1e4d

Commits on May 2, 2023

update megatron convert script
void-main
committed
da2ad14
Copy full SHA for da2ad14

Commits on May 1, 2023

Merge branch 'main' of https://github.com/void-main/FasterTransformer into main
void-main
committed
d752088
Copy full SHA for d752088
load layernorm_eps from config; change cb default to 5
void-main
committed
f6cf9da
Copy full SHA for f6cf9da
bugfix
void-main
committed
40fbe48
Copy full SHA for 40fbe48
Fix/gpt early stop (#584)
Show description for c6e8f60
byshiue
authored
c6e8f60
Copy full SHA for c6e8f60

Commits on Apr 30, 2023

Merge branch 'main' of https://github.com/void-main/FasterTransformer into main
void-main
committed
694faec
Copy full SHA for 694faec
support bf16 conversion with bfloat 16 numpy ext
void-main
committed
8ddac81
Copy full SHA for 8ddac81

Commits on Apr 29, 2023

make sure examples work for bf16
void-main
committed
9aee02e
Copy full SHA for 9aee02e
add bf16
void-main
committed
95afed4
Copy full SHA for 95afed4
donot callback too frequnetly
void-main
committed
cf1b9b1
Copy full SHA for cf1b9b1

Commits on Apr 28, 2023

Update src/fastertransformer/triton_backend/llama/LlamaTritonModelInstance.h
Show description for d65adf1

void-main
and
bwasti
authored
d65adf1
Copy full SHA for d65adf1
add megatron llama convert
void-main
committed
0379cc5
Copy full SHA for 0379cc5

Commits on Apr 25, 2023

bugfix
void-main
committed
bfeebef
Copy full SHA for bfeebef
revert bf16
void-main
committed
9820565
Copy full SHA for 9820565
revert bf16
void-main
committed
7a72ca3
Copy full SHA for 7a72ca3
register RMS for bf16
void-main
committed
a6d51ec
Copy full SHA for a6d51ec
add triton model for streaming callback
void-main
committed
4bc97c3
Copy full SHA for 4bc97c3
add bf16 support
void-main
committed
91989cb
Copy full SHA for 91989cb

Commits on Apr 24, 2023

fix llama rms ln
void-main
committed
ce8700f
Copy full SHA for ce8700f
perf(bloom): improve performance of huggingface_bloom_convert.py, decrease the time cost and the mem using (#568)
Show description for 19b2956

Yangruipis
and
r.yang
authored
19b2956
Copy full SHA for 19b2956
make the code work :yay:
void-main
committed
a32fc1d
Copy full SHA for a32fc1d