huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-09-12 04:44:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(fp8): add support for fbgemm

Browse Source
This commit is contained in:
OlivierDehaene 2024-07-18 12:40:57 +02:00
parent 11123a8e99
commit 27084bbfd3
No known key found for this signature in database
GPG Key ID: BB104D67809DA93C
7 changed files with 756 additions and 644 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
server/Makefile
View File

@ -36,5 +36,6 @@ run-dev:
SAFETENSORS_FAST_GPU=1 python -m torch.distributed.run --nproc_per_node=2 text_generation_server/cli.py serve bigscience/bloom-560m --sharded SAFETENSORS_FAST_GPU=1 python -m torch.distributed.run --nproc_per_node=2 text_generation_server/cli.py serve bigscience/bloom-560m --sharded
export-requirements: export-requirements:
poetry export -o requirements_cuda.txt --without-hashes --with cuda poetry export -o requirements_cuda.txt --without-hashes -E cuda
poetry export -o requirements_rocm.txt --without-hashes poetry export -o requirements_rocm.txt --without-hashes
poetry export -o requirements_intel.txt --without-hashes

1206
server/poetry.lock generated
View File

File diff suppressed because it is too large Load Diff

2
server/pyproject.toml
View File

@ -37,9 +37,11 @@ pillow = "^10.0.0"
outlines= { version = "^0.0.34", optional = true } outlines= { version = "^0.0.34", optional = true }
prometheus-client = "^0.20.0" prometheus-client = "^0.20.0"
py-cpuinfo = "^9.0.0" py-cpuinfo = "^9.0.0"
fbgemm-gpu = { version = "0.8.0rc4", optional = true }
[tool.poetry.extras] [tool.poetry.extras]
torch = ["torch"] torch = ["torch"]
cuda = ["fbgemm-gpu"]
accelerate = ["accelerate"] accelerate = ["accelerate"]
bnb = ["bitsandbytes"] bnb = ["bitsandbytes"]
peft = ["peft"] peft = ["peft"]

49
server/requirements_cuda.txt
View File

@ -1,48 +1,51 @@
backoff==2.2.1 ; python_version >= "3.9" and python_version < "3.13" certifi==2024.7.4 ; python_version >= "3.9" and python_version < "3.13"
certifi==2024.2.2 ; python_version >= "3.9" and python_version < "3.13"
charset-normalizer==3.3.2 ; python_version >= "3.9" and python_version < "3.13" charset-normalizer==3.3.2 ; python_version >= "3.9" and python_version < "3.13"
click==8.1.7 ; python_version >= "3.9" and python_version < "3.13" click==8.1.7 ; python_version >= "3.9" and python_version < "3.13"
colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.13" and (sys_platform == "win32" or platform_system == "Windows") colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.13" and (sys_platform == "win32" or platform_system == "Windows")
deprecated==1.2.14 ; python_version >= "3.9" and python_version < "3.13" deprecated==1.2.14 ; python_version >= "3.9" and python_version < "3.13"
einops==0.6.1 ; python_version >= "3.9" and python_version < "3.13" einops==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
filelock==3.14.0 ; python_version >= "3.9" and python_version < "3.13" fbgemm-gpu==0.8.0rc4 ; python_version >= "3.9" and python_version < "3.13"
filelock==3.15.4 ; python_version >= "3.9" and python_version < "3.13"
fsspec==2024.5.0 ; python_version >= "3.9" and python_version < "3.13" fsspec==2024.5.0 ; python_version >= "3.9" and python_version < "3.13"
googleapis-common-protos==1.63.0 ; python_version >= "3.9" and python_version < "3.13" googleapis-common-protos==1.63.2 ; python_version >= "3.9" and python_version < "3.13"
grpc-interceptor==0.15.4 ; python_version >= "3.9" and python_version < "3.13" grpc-interceptor==0.15.4 ; python_version >= "3.9" and python_version < "3.13"
grpcio-reflection==1.62.2 ; python_version >= "3.9" and python_version < "3.13" grpcio-reflection==1.62.2 ; python_version >= "3.9" and python_version < "3.13"
grpcio-status==1.62.2 ; python_version >= "3.9" and python_version < "3.13" grpcio-status==1.62.2 ; python_version >= "3.9" and python_version < "3.13"
grpcio==1.64.0 ; python_version >= "3.9" and python_version < "3.13" grpcio==1.65.1 ; python_version >= "3.9" and python_version < "3.13"
hf-transfer==0.1.6 ; python_version >= "3.9" and python_version < "3.13" hf-transfer==0.1.6 ; python_version >= "3.9" and python_version < "3.13"
huggingface-hub==0.23.1 ; python_version >= "3.9" and python_version < "3.13" huggingface-hub==0.23.5 ; python_version >= "3.9" and python_version < "3.13"
idna==3.7 ; python_version >= "3.9" and python_version < "3.13" idna==3.7 ; python_version >= "3.9" and python_version < "3.13"
importlib-metadata==7.1.0 ; python_version >= "3.9" and python_version < "3.13"
loguru==0.6.0 ; python_version >= "3.9" and python_version < "3.13" loguru==0.6.0 ; python_version >= "3.9" and python_version < "3.13"
numpy==1.26.4 ; python_version >= "3.9" and python_version < "3.13" numpy==2.0.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-api==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-api==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp-proto-grpc==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-common==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp-proto-http==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-grpc==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-http==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-instrumentation-grpc==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-instrumentation==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-instrumentation-grpc==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-proto==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-instrumentation==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-sdk==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-proto==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-semantic-conventions==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-sdk==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
packaging==24.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-semantic-conventions==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
pillow==10.3.0 ; python_version >= "3.9" and python_version < "3.13" packaging==24.1 ; python_version >= "3.9" and python_version < "3.13"
pillow==10.4.0 ; python_version >= "3.9" and python_version < "3.13"
prometheus-client==0.20.0 ; python_version >= "3.9" and python_version < "3.13" prometheus-client==0.20.0 ; python_version >= "3.9" and python_version < "3.13"
protobuf==4.25.3 ; python_version >= "3.9" and python_version < "3.13" protobuf==4.25.3 ; python_version >= "3.9" and python_version < "3.13"
py-cpuinfo==9.0.0 ; python_version >= "3.9" and python_version < "3.13" py-cpuinfo==9.0.0 ; python_version >= "3.9" and python_version < "3.13"
pyyaml==6.0.1 ; python_version >= "3.9" and python_version < "3.13" pyyaml==6.0.1 ; python_version >= "3.9" and python_version < "3.13"
regex==2024.5.15 ; python_version >= "3.9" and python_version < "3.13" regex==2024.5.15 ; python_version >= "3.9" and python_version < "3.13"
requests==2.32.2 ; python_version >= "3.9" and python_version < "3.13" requests==2.32.3 ; python_version >= "3.9" and python_version < "3.13"
safetensors==0.4.3 ; python_version >= "3.9" and python_version < "3.13" safetensors==0.4.3 ; python_version >= "3.9" and python_version < "3.13"
scipy==1.13.1 ; python_version >= "3.9" and python_version < "3.13" scipy==1.13.1 ; python_version >= "3.9" and python_version < "3.13"
sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.13" sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.13"
setuptools==70.0.0 ; python_version >= "3.9" and python_version < "3.13" setuptools==71.0.1 ; python_version >= "3.9" and python_version < "3.13"
tokenizers==0.19.1 ; python_version >= "3.9" and python_version < "3.13" tokenizers==0.19.1 ; python_version >= "3.9" and python_version < "3.13"
tqdm==4.66.4 ; python_version >= "3.9" and python_version < "3.13" tqdm==4.66.4 ; python_version >= "3.9" and python_version < "3.13"
transformers==4.41.1 ; python_version >= "3.9" and python_version < "3.13" transformers==4.41.2 ; python_version >= "3.9" and python_version < "3.13"
typer==0.6.1 ; python_version >= "3.9" and python_version < "3.13" typer==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
typing-extensions==4.12.0 ; python_version >= "3.9" and python_version < "3.13" typing-extensions==4.12.2 ; python_version >= "3.9" and python_version < "3.13"
urllib3==2.2.1 ; python_version >= "3.9" and python_version < "3.13" urllib3==2.2.2 ; python_version >= "3.9" and python_version < "3.13"
win32-setctime==1.1.0 ; python_version >= "3.9" and python_version < "3.13" and sys_platform == "win32" win32-setctime==1.1.0 ; python_version >= "3.9" and python_version < "3.13" and sys_platform == "win32"
wrapt==1.16.0 ; python_version >= "3.9" and python_version < "3.13" wrapt==1.16.0 ; python_version >= "3.9" and python_version < "3.13"
zipp==3.19.2 ; python_version >= "3.9" and python_version < "3.13"

48
server/requirements_intel.txt
View File

@ -1,48 +1,50 @@
backoff==2.2.1 ; python_version >= "3.9" and python_version < "3.13" certifi==2024.7.4 ; python_version >= "3.9" and python_version < "3.13"
certifi==2024.2.2 ; python_version >= "3.9" and python_version < "3.13"
charset-normalizer==3.3.2 ; python_version >= "3.9" and python_version < "3.13" charset-normalizer==3.3.2 ; python_version >= "3.9" and python_version < "3.13"
click==8.1.7 ; python_version >= "3.9" and python_version < "3.13" click==8.1.7 ; python_version >= "3.9" and python_version < "3.13"
colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.13" and (sys_platform == "win32" or platform_system == "Windows") colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.13" and (sys_platform == "win32" or platform_system == "Windows")
deprecated==1.2.14 ; python_version >= "3.9" and python_version < "3.13" deprecated==1.2.14 ; python_version >= "3.9" and python_version < "3.13"
einops==0.6.1 ; python_version >= "3.9" and python_version < "3.13" einops==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
filelock==3.14.0 ; python_version >= "3.9" and python_version < "3.13" filelock==3.15.4 ; python_version >= "3.9" and python_version < "3.13"
fsspec==2024.5.0 ; python_version >= "3.9" and python_version < "3.13" fsspec==2024.5.0 ; python_version >= "3.9" and python_version < "3.13"
googleapis-common-protos==1.63.0 ; python_version >= "3.9" and python_version < "3.13" googleapis-common-protos==1.63.2 ; python_version >= "3.9" and python_version < "3.13"
grpc-interceptor==0.15.4 ; python_version >= "3.9" and python_version < "3.13" grpc-interceptor==0.15.4 ; python_version >= "3.9" and python_version < "3.13"
grpcio-reflection==1.62.2 ; python_version >= "3.9" and python_version < "3.13" grpcio-reflection==1.62.2 ; python_version >= "3.9" and python_version < "3.13"
grpcio-status==1.62.2 ; python_version >= "3.9" and python_version < "3.13" grpcio-status==1.62.2 ; python_version >= "3.9" and python_version < "3.13"
grpcio==1.64.0 ; python_version >= "3.9" and python_version < "3.13" grpcio==1.65.1 ; python_version >= "3.9" and python_version < "3.13"
hf-transfer==0.1.6 ; python_version >= "3.9" and python_version < "3.13" hf-transfer==0.1.6 ; python_version >= "3.9" and python_version < "3.13"
huggingface-hub==0.23.1 ; python_version >= "3.9" and python_version < "3.13" huggingface-hub==0.23.5 ; python_version >= "3.9" and python_version < "3.13"
idna==3.7 ; python_version >= "3.9" and python_version < "3.13" idna==3.7 ; python_version >= "3.9" and python_version < "3.13"
importlib-metadata==7.1.0 ; python_version >= "3.9" and python_version < "3.13"
loguru==0.6.0 ; python_version >= "3.9" and python_version < "3.13" loguru==0.6.0 ; python_version >= "3.9" and python_version < "3.13"
numpy==1.26.4 ; python_version >= "3.9" and python_version < "3.13" numpy==2.0.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-api==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-api==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp-proto-grpc==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-common==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp-proto-http==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-grpc==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-http==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-instrumentation-grpc==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-instrumentation==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-instrumentation-grpc==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-proto==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-instrumentation==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-sdk==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-proto==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-semantic-conventions==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-sdk==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
packaging==24.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-semantic-conventions==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
pillow==10.3.0 ; python_version >= "3.9" and python_version < "3.13" packaging==24.1 ; python_version >= "3.9" and python_version < "3.13"
pillow==10.4.0 ; python_version >= "3.9" and python_version < "3.13"
prometheus-client==0.20.0 ; python_version >= "3.9" and python_version < "3.13" prometheus-client==0.20.0 ; python_version >= "3.9" and python_version < "3.13"
protobuf==4.25.3 ; python_version >= "3.9" and python_version < "3.13" protobuf==4.25.3 ; python_version >= "3.9" and python_version < "3.13"
py-cpuinfo==9.0.0 ; python_version >= "3.9" and python_version < "3.13" py-cpuinfo==9.0.0 ; python_version >= "3.9" and python_version < "3.13"
pyyaml==6.0.1 ; python_version >= "3.9" and python_version < "3.13" pyyaml==6.0.1 ; python_version >= "3.9" and python_version < "3.13"
regex==2024.5.15 ; python_version >= "3.9" and python_version < "3.13" regex==2024.5.15 ; python_version >= "3.9" and python_version < "3.13"
requests==2.32.2 ; python_version >= "3.9" and python_version < "3.13" requests==2.32.3 ; python_version >= "3.9" and python_version < "3.13"
safetensors==0.4.3 ; python_version >= "3.9" and python_version < "3.13" safetensors==0.4.3 ; python_version >= "3.9" and python_version < "3.13"
scipy==1.13.1 ; python_version >= "3.9" and python_version < "3.13" scipy==1.13.1 ; python_version >= "3.9" and python_version < "3.13"
sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.13" sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.13"
setuptools==69.5.1 ; python_version >= "3.9" and python_version < "3.13" setuptools==71.0.1 ; python_version >= "3.9" and python_version < "3.13"
tokenizers==0.19.1 ; python_version >= "3.9" and python_version < "3.13" tokenizers==0.19.1 ; python_version >= "3.9" and python_version < "3.13"
tqdm==4.66.4 ; python_version >= "3.9" and python_version < "3.13" tqdm==4.66.4 ; python_version >= "3.9" and python_version < "3.13"
transformers==4.41.1 ; python_version >= "3.9" and python_version < "3.13" transformers==4.41.2 ; python_version >= "3.9" and python_version < "3.13"
typer==0.6.1 ; python_version >= "3.9" and python_version < "3.13" typer==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
typing-extensions==4.12.0 ; python_version >= "3.9" and python_version < "3.13" typing-extensions==4.12.2 ; python_version >= "3.9" and python_version < "3.13"
urllib3==2.2.1 ; python_version >= "3.9" and python_version < "3.13" urllib3==2.2.2 ; python_version >= "3.9" and python_version < "3.13"
win32-setctime==1.1.0 ; python_version >= "3.9" and python_version < "3.13" and sys_platform == "win32" win32-setctime==1.1.0 ; python_version >= "3.9" and python_version < "3.13" and sys_platform == "win32"
wrapt==1.16.0 ; python_version >= "3.9" and python_version < "3.13" wrapt==1.16.0 ; python_version >= "3.9" and python_version < "3.13"
zipp==3.19.2 ; python_version >= "3.9" and python_version < "3.13"

48
server/requirements_rocm.txt
View File

@ -1,48 +1,50 @@
backoff==2.2.1 ; python_version >= "3.9" and python_version < "3.13" certifi==2024.7.4 ; python_version >= "3.9" and python_version < "3.13"
certifi==2024.2.2 ; python_version >= "3.9" and python_version < "3.13"
charset-normalizer==3.3.2 ; python_version >= "3.9" and python_version < "3.13" charset-normalizer==3.3.2 ; python_version >= "3.9" and python_version < "3.13"
click==8.1.7 ; python_version >= "3.9" and python_version < "3.13" click==8.1.7 ; python_version >= "3.9" and python_version < "3.13"
colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.13" and (sys_platform == "win32" or platform_system == "Windows") colorama==0.4.6 ; python_version >= "3.9" and python_version < "3.13" and (sys_platform == "win32" or platform_system == "Windows")
deprecated==1.2.14 ; python_version >= "3.9" and python_version < "3.13" deprecated==1.2.14 ; python_version >= "3.9" and python_version < "3.13"
einops==0.6.1 ; python_version >= "3.9" and python_version < "3.13" einops==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
filelock==3.14.0 ; python_version >= "3.9" and python_version < "3.13" filelock==3.15.4 ; python_version >= "3.9" and python_version < "3.13"
fsspec==2024.5.0 ; python_version >= "3.9" and python_version < "3.13" fsspec==2024.5.0 ; python_version >= "3.9" and python_version < "3.13"
googleapis-common-protos==1.63.0 ; python_version >= "3.9" and python_version < "3.13" googleapis-common-protos==1.63.2 ; python_version >= "3.9" and python_version < "3.13"
grpc-interceptor==0.15.4 ; python_version >= "3.9" and python_version < "3.13" grpc-interceptor==0.15.4 ; python_version >= "3.9" and python_version < "3.13"
grpcio-reflection==1.62.2 ; python_version >= "3.9" and python_version < "3.13" grpcio-reflection==1.62.2 ; python_version >= "3.9" and python_version < "3.13"
grpcio-status==1.62.2 ; python_version >= "3.9" and python_version < "3.13" grpcio-status==1.62.2 ; python_version >= "3.9" and python_version < "3.13"
grpcio==1.64.0 ; python_version >= "3.9" and python_version < "3.13" grpcio==1.65.1 ; python_version >= "3.9" and python_version < "3.13"
hf-transfer==0.1.6 ; python_version >= "3.9" and python_version < "3.13" hf-transfer==0.1.6 ; python_version >= "3.9" and python_version < "3.13"
huggingface-hub==0.23.1 ; python_version >= "3.9" and python_version < "3.13" huggingface-hub==0.23.5 ; python_version >= "3.9" and python_version < "3.13"
idna==3.7 ; python_version >= "3.9" and python_version < "3.13" idna==3.7 ; python_version >= "3.9" and python_version < "3.13"
importlib-metadata==7.1.0 ; python_version >= "3.9" and python_version < "3.13"
loguru==0.6.0 ; python_version >= "3.9" and python_version < "3.13" loguru==0.6.0 ; python_version >= "3.9" and python_version < "3.13"
numpy==1.26.4 ; python_version >= "3.9" and python_version < "3.13" numpy==2.0.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-api==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-api==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp-proto-grpc==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-common==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp-proto-http==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-grpc==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-exporter-otlp==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp-proto-http==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-instrumentation-grpc==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-exporter-otlp==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-instrumentation==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-instrumentation-grpc==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-proto==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-instrumentation==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-sdk==1.15.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-proto==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
opentelemetry-semantic-conventions==0.36b0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-sdk==1.25.0 ; python_version >= "3.9" and python_version < "3.13"
packaging==24.0 ; python_version >= "3.9" and python_version < "3.13" opentelemetry-semantic-conventions==0.46b0 ; python_version >= "3.9" and python_version < "3.13"
pillow==10.3.0 ; python_version >= "3.9" and python_version < "3.13" packaging==24.1 ; python_version >= "3.9" and python_version < "3.13"
pillow==10.4.0 ; python_version >= "3.9" and python_version < "3.13"
prometheus-client==0.20.0 ; python_version >= "3.9" and python_version < "3.13" prometheus-client==0.20.0 ; python_version >= "3.9" and python_version < "3.13"
protobuf==4.25.3 ; python_version >= "3.9" and python_version < "3.13" protobuf==4.25.3 ; python_version >= "3.9" and python_version < "3.13"
py-cpuinfo==9.0.0 ; python_version >= "3.9" and python_version < "3.13" py-cpuinfo==9.0.0 ; python_version >= "3.9" and python_version < "3.13"
pyyaml==6.0.1 ; python_version >= "3.9" and python_version < "3.13" pyyaml==6.0.1 ; python_version >= "3.9" and python_version < "3.13"
regex==2024.5.15 ; python_version >= "3.9" and python_version < "3.13" regex==2024.5.15 ; python_version >= "3.9" and python_version < "3.13"
requests==2.32.2 ; python_version >= "3.9" and python_version < "3.13" requests==2.32.3 ; python_version >= "3.9" and python_version < "3.13"
safetensors==0.4.3 ; python_version >= "3.9" and python_version < "3.13" safetensors==0.4.3 ; python_version >= "3.9" and python_version < "3.13"
scipy==1.13.1 ; python_version >= "3.9" and python_version < "3.13" scipy==1.13.1 ; python_version >= "3.9" and python_version < "3.13"
sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.13" sentencepiece==0.1.99 ; python_version >= "3.9" and python_version < "3.13"
setuptools==70.0.0 ; python_version >= "3.9" and python_version < "3.13" setuptools==71.0.1 ; python_version >= "3.9" and python_version < "3.13"
tokenizers==0.19.1 ; python_version >= "3.9" and python_version < "3.13" tokenizers==0.19.1 ; python_version >= "3.9" and python_version < "3.13"
tqdm==4.66.4 ; python_version >= "3.9" and python_version < "3.13" tqdm==4.66.4 ; python_version >= "3.9" and python_version < "3.13"
transformers==4.41.1 ; python_version >= "3.9" and python_version < "3.13" transformers==4.41.2 ; python_version >= "3.9" and python_version < "3.13"
typer==0.6.1 ; python_version >= "3.9" and python_version < "3.13" typer==0.6.1 ; python_version >= "3.9" and python_version < "3.13"
typing-extensions==4.12.0 ; python_version >= "3.9" and python_version < "3.13" typing-extensions==4.12.2 ; python_version >= "3.9" and python_version < "3.13"
urllib3==2.2.1 ; python_version >= "3.9" and python_version < "3.13" urllib3==2.2.2 ; python_version >= "3.9" and python_version < "3.13"
win32-setctime==1.1.0 ; python_version >= "3.9" and python_version < "3.13" and sys_platform == "win32" win32-setctime==1.1.0 ; python_version >= "3.9" and python_version < "3.13" and sys_platform == "win32"
wrapt==1.16.0 ; python_version >= "3.9" and python_version < "3.13" wrapt==1.16.0 ; python_version >= "3.9" and python_version < "3.13"
zipp==3.19.2 ; python_version >= "3.9" and python_version < "3.13"

44
server/text_generation_server/layers/fp8.py
View File

@ -4,6 +4,13 @@ import torch
from text_generation_server.utils.import_utils import SYSTEM from text_generation_server.utils.import_utils import SYSTEM
from text_generation_server.utils.weights import Weight from text_generation_server.utils.weights import Weight
try:
import fbgemm_gpu.experimental.gen_ai
HAS_FBGEMM = True
except (ImportError, ModuleNotFoundError):
HAS_FBGEMM = False
def get_fp8_linear() -> torch.nn.Module: def get_fp8_linear() -> torch.nn.Module:
""" """
@ -21,12 +28,20 @@ def get_fp8_linear() -> torch.nn.Module:
return Fp8Linear return Fp8Linear
def fp8_quantize(weight, qdtype=torch.float8_e4m3fn): def fp8_quantize(weight, scale_upper_bound=None, qdtype=torch.float8_e4m3fn):
device = weight.device if HAS_FBGEMM:
if scale_upper_bound.device != weight.device:
scale_upper_bound = scale_upper_bound.to(weight.device)
qweight, scale = torch.ops.fbgemm.quantize_fp8_per_row(
weight, bs=None, scale_ub=scale_upper_bound, output_dtype=qdtype
)
return qweight, scale
# weight, scale = quant_weights(weight, torch.int8, False) # weight, scale = quant_weights(weight, torch.int8, False)
finfo = torch.finfo(qdtype) finfo = torch.finfo(qdtype)
# Calculate the scale as dtype max divided by absmax # Calculate the scale as dtype max divided by absmax
scale = finfo.max / weight.abs().max().clamp(min=1e-12) scale = finfo.max / weight.abs().max().clamp(min=1e-12, max=scale_upper_bound)
# scale and clamp the tensor to bring it to # scale and clamp the tensor to bring it to
# the representative range of float8 data type # the representative range of float8 data type
# (as default cast is unsaturated) # (as default cast is unsaturated)
@ -59,6 +74,29 @@ class Fp8Linear(torch.nn.Module):
self.bias = bias if bias is not None else None self.bias = bias if bias is not None else None
def forward(self, input: torch.Tensor) -> torch.Tensor: def forward(self, input: torch.Tensor) -> torch.Tensor:
if HAS_FBGEMM:
global default_activation_scale_upper_bound
device = input.device
if default_activation_scale_upper_bound.device != device:
default_activation_scale_upper_bound = (
default_activation_scale_upper_bound.to(device)
)
qinput, scale = fp8_quantize(
input, scale_upper_bound=default_activation_scale_upper_bound
)
y = torch.ops.fbgemm.f8f8bf16_rowwise(
qinput,
self.weight,
scale,
self.scale,
use_fast_accum=True,
bias=self.bias,
)
return y.to(self.dtype)
qinput, scale = fp8_quantize(input) qinput, scale = fp8_quantize(input)
output, _ = torch._scaled_mm( output, _ = torch._scaled_mm(
qinput, qinput,