Spaces:
Paused
Paused
Update Dockerfile
Browse files- Dockerfile +19 -17
Dockerfile
CHANGED
@@ -1,26 +1,28 @@
|
|
1 |
-
FROM nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu22.04
|
2 |
|
3 |
-
|
4 |
-
|
5 |
-
ENV OMP_NUM_THREADS=1
|
6 |
|
7 |
-
#
|
8 |
-
ENV
|
|
|
9 |
|
10 |
-
# https://github.com/vllm-project/vllm/blob/v0.10.0/docs/getting_started/
|
11 |
-
#
|
12 |
-
ENV VLLM_ATTENTION_BACKEND=FLASH_ATTN
|
13 |
|
14 |
-
|
15 |
-
|
|
|
16 |
|
17 |
-
#
|
18 |
-
RUN pip3 install
|
19 |
|
20 |
-
|
21 |
-
|
22 |
|
23 |
-
RUN
|
|
|
|
|
|
|
24 |
|
25 |
# Download at build time,
|
26 |
# to ensure during restart we won't have to wait for the download from HF (only wait for docker pull).
|
@@ -39,7 +41,7 @@ RUN mkdir -p /tmp/.cache/huggingface
|
|
39 |
EXPOSE 7860
|
40 |
|
41 |
# Export for runtime environment
|
42 |
-
CMD
|
43 |
--model "meta-llama/Llama-3.2-3B-Instruct" \
|
44 |
--task generate \
|
45 |
--revision "0cb88a4f764b7a12671c53f0838cd831a0843b95" \
|
|
|
|
|
1 |
|
2 |
+
FROM vllm/vllm-openai:v0.10.0
|
3 |
+
# FROM nvidia/cuda:11.7.1-cudnn8-runtime-ubuntu22.04
|
|
|
4 |
|
5 |
+
# ENV VLLM_LOGGING_LEVEL=DEBUG
|
6 |
+
# ENV HF_HOME=/tmp/.cache/huggingface
|
7 |
+
# ENV OMP_NUM_THREADS=1
|
8 |
|
9 |
+
# # https://github.com/vllm-project/vllm/blob/v0.10.0/docs/getting_started/installation/gpu/rocm.inc.md?plain=1#L124
|
10 |
+
# ENV VLLM_USE_TRITON_FLASH_ATTN=0
|
|
|
11 |
|
12 |
+
# # https://github.com/vllm-project/vllm/blob/v0.10.0/docs/getting_started/quickstart.md?plain=1#L213
|
13 |
+
# # `FLASH_ATTN` or `FLASHINFER` or `XFORMERS`.
|
14 |
+
# ENV VLLM_ATTENTION_BACKEND=FLASH_ATTN
|
15 |
|
16 |
+
# RUN apt-get update && apt-get install -y python3 python3-pip git
|
17 |
+
# RUN pip3 install --upgrade pip
|
18 |
|
19 |
+
# # Install vLLM
|
20 |
+
# RUN pip3 install vllm==0.10.0 --extra-index-url https://download.pytorch.org/whl/cu113
|
21 |
|
22 |
+
# RUN useradd -m appuser
|
23 |
+
# USER appuser
|
24 |
+
|
25 |
+
# RUN mkdir -p /tmp/.cache/huggingface
|
26 |
|
27 |
# Download at build time,
|
28 |
# to ensure during restart we won't have to wait for the download from HF (only wait for docker pull).
|
|
|
41 |
EXPOSE 7860
|
42 |
|
43 |
# Export for runtime environment
|
44 |
+
CMD vllm serve \
|
45 |
--model "meta-llama/Llama-3.2-3B-Instruct" \
|
46 |
--task generate \
|
47 |
--revision "0cb88a4f764b7a12671c53f0838cd831a0843b95" \
|