Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add trtllm container update #2191

Merged
merged 2 commits into from
Jul 18, 2024
Merged

add trtllm container update #2191

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f54568a
update installation
Jul 17, 2024
b864e6b
use cloudfront
Jul 18, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 8 additions & 8 deletions serving/docker/tensorrt-llm.Dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,20 +15,20 @@ ARG cuda_version=cu124
ARG python_version=3.10
ARG TORCH_VERSION=2.3.1
ARG djl_version=0.29.0~SNAPSHOT
ARG transformers_version=4.40.0
ARG accelerate_version=0.29.3
ARG tensorrtlibs_version=10.0.1
ARG transformers_version=4.42.4
ARG accelerate_version=0.32.1
ARG tensorrtlibs_version=10.1.0
# %2B is the url escape for the '+' character
ARG trtllm_toolkit_version=0.10.0%2Bnightly
ARG trtllm_version=v0.10.0
ARG trtllm_toolkit_version=0.11.0%2Bnightly
ARG trtllm_version=v0.11.0
ARG cuda_python_version=12.4
ARG peft_version=0.10.0
ARG triton_version=r24.04
ARG trtllm_toolkit_wheel="https://publish.djl.ai/tensorrt-llm/toolkit/tensorrt_llm_toolkit-${trtllm_toolkit_version}-py3-none-any.whl"
ARG trtllm_wheel="https://djl-ai.s3.amazonaws.com/publish/tensorrt-llm/${trtllm_version}/tensorrt_llm-0.10.0-cp310-cp310-linux_x86_64.whl"
ARG trtllm_wheel="https://publish.djl.ai/tensorrt-llm/${trtllm_version}/tensorrt_llm-0.11.0-cp310-cp310-linux_x86_64.whl"
Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@ydm-amazon this will allow us to use cloudfront (an AWS services) that do global CDN on the artifacts we have and provide 10x speed up on downloading. Using raw HTTP url for S3 will be slower than that.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's awesome

ARG triton_toolkit_wheel="https://publish.djl.ai/tritonserver/${triton_version}/tritontoolkit-24.4-py310-none-any.whl"
ARG pydantic_version=2.6.1
ARG modelopt_version=0.11.2
ARG modelopt_version=0.13.1
ARG janus_version=1.0.0
ARG pynvml_verison=11.5.0
ARG numpy_version=1.26.4
Expand Down Expand Up @@ -68,7 +68,7 @@ COPY distribution[s]/ ./
RUN mv *.deb djl-serving_all.deb || true

# Install CUDNN 8
RUN apt-get update && apt-get install -y --no-install-recommends libcudnn8 && rm -rf /var/lib/apt/lists/*
RUN apt-get update && apt-get install -y --no-install-recommends libcudnn9-cuda-12 && rm -rf /var/lib/apt/lists/*
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why not just use 12.4.1-cudnn-devel-ubuntu22.04 image directly?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

it also installed libcudnn9-dev-cuda-12, which is around 1GB we don't need that


# Install OpenMPI and other deps
ARG DEBIAN_FRONTEND=noninteractive
Expand Down
6 changes: 0 additions & 6 deletions tests/integration/llm/client.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -503,12 +503,6 @@ def get_model_name():
"seq_length": [256],
"tokenizer": "TheBloke/Llama-2-13B-fp16"
},
"falcon-7b": {
"max_memory_per_gpu": [22.0],
"batch_size": [1, 4],
"seq_length": [256],
"tokenizer": "tiiuae/falcon-7b"
},
"llama2-7b-smoothquant": {
"max_memory_per_gpu": [22.0],
"batch_size": [1, 4],
Expand Down
9 changes: 0 additions & 9 deletions tests/integration/llm/prepare.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -784,15 +784,6 @@
"option.rolling_batch": "trtllm",
"option.output_formatter": "jsonlines",
},
"falcon-7b": {
"option.model_id": "s3://djl-llm/triton/0.10.0/falcon-7b-tp1-bs16/",
"option.tensor_parallel_degree": 1,
"option.max_input_len": 1024,
"option.max_output_len": 512,
"option.max_rolling_batch_size": 16,
"option.rolling_batch": "auto",
"option.output_formatter": "jsonlines",
},
"llama2-7b-smoothquant": {
"option.model_id": "s3://djl-llm/meta-llama-Llama-2-7b-chat-hf/",
"option.tensor_parallel_degree": 4,
Expand Down
6 changes: 0 additions & 6 deletions tests/integration/tests.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -266,12 +266,6 @@ def test_llama2_13b_tp4(self):
r.launch("CUDA_VISIBLE_DEVICES=0,1,2,3")
client.run("trtllm llama2-13b".split())

def test_falcon_triton(self):
with Runner('tensorrt-llm', 'falcon-7b') as r:
prepare.build_trtllm_handler_model("falcon-7b")
r.launch("CUDA_VISIBLE_DEVICES=0")
client.run("trtllm falcon-7b".split())

def test_internlm_7b(self):
with Runner('tensorrt-llm', 'internlm-7b') as r:
prepare.build_trtllm_handler_model("internlm-7b")
Expand Down
Loading