Spaces:

anews9340
/

llama.cpp

Paused

anews9340 commited on Mar 2

Commit

c4c2b5b

verified ·

1 Parent(s): 7e4778c

Update Dockerfile

Files changed (1) hide show

Dockerfile CHANGED Viewed

@@ -45,11 +45,11 @@ RUN git clone --depth 1 https://github.com/ggerganov/llama.cpp.git . && \
 ENTRYPOINT ["./build/bin/llama-server"]
 CMD [ \
-    "-hf", "unsloth/Qwen3.5-9B-GGUF:Q8_0", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "-t", "8", \
     "-c", "4096", \
-    "--flash-attn", "true"
-    "--no-mmap" \
     ]

 ENTRYPOINT ["./build/bin/llama-server"]
 CMD [ \
+    "-hf", "unsloth/Qwen3.5-4B-GGUF:Q8_0", \
     "--host", "0.0.0.0", \
     "--port", "7860", \
     "-t", "8", \
     "-c", "4096", \
+    "--flash-attn", "true", \
+    "--no-mmap"
     ]