muryshev commited on
Commit
911955a
·
1 Parent(s): fa4f8bb

Update Dockerfile

Browse files
Files changed (1) hide show
  1. Dockerfile +3 -3
Dockerfile CHANGED
@@ -21,12 +21,12 @@ ENV LLAMA_CUBLAS=1
21
 
22
  RUN mkdir build && \
23
  cd build && \
24
- # cmake .. #-DLLAMA_CUBLAS=ON && \
25
  cmake .. && \
26
  cmake --build . --config Release
27
 
28
  WORKDIR /data
29
- RUN wget https://huggingface.co/IlyaGusev/saiga2_13b_gguf/resolve/main/model-q8_0.gguf -nv -O model.gguf
30
 
31
  FROM ${BASE_CUDA_RUN_CONTAINER} as runtime
32
 
@@ -39,4 +39,4 @@ COPY --from=build /data/model.gguf /data/model.gguf
39
  WORKDIR /app
40
  EXPOSE 7860
41
 
42
- CMD ./server -m /data/model.gguf -c 2048 --port 7860 --host 0.0.0.0
 
21
 
22
  RUN mkdir build && \
23
  cd build && \
24
+ cmake .. #-DLLAMA_CUBLAS=ON && \
25
  cmake .. && \
26
  cmake --build . --config Release
27
 
28
  WORKDIR /data
29
+ RUN wget https://huggingface.co/IlyaGusev/saiga2_70b_gguf/resolve/main/ggml-model-q4_1.gguf -nv -O model.gguf
30
 
31
  FROM ${BASE_CUDA_RUN_CONTAINER} as runtime
32
 
 
39
  WORKDIR /app
40
  EXPOSE 7860
41
 
42
+ CMD ./server -m /data/model.gguf -c 4096 --port 7860 --host 0.0.0.0