Spaces:
Paused
Paused
Create Dockerfile
Browse files- Dockerfile +16 -0
Dockerfile
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
FROM ghcr.io/ggml-org/llama.cpp:server
|
| 2 |
+
|
| 3 |
+
# Forces the server to use a writeable temporary directory for Hugging Face compatibility
|
| 4 |
+
ENV HOME=/tmp
|
| 5 |
+
|
| 6 |
+
# Expose the mandatory port for Hugging Face routing
|
| 7 |
+
EXPOSE 7860
|
| 8 |
+
|
| 9 |
+
# Run the official, ultra-optimized C++ server binary directly
|
| 10 |
+
CMD [ \
|
| 11 |
+
"--host", "0.0.0.0", \
|
| 12 |
+
"--port", "7860", \
|
| 13 |
+
"--model", "https://huggingface.co/bartowski/google_gemma-3-4b-it-GGUF/resolve/main/google_gemma-3-4b-it-Q4_K_M.gguf", \
|
| 14 |
+
"--ctx-size", "2048", \
|
| 15 |
+
"--threads", "2" \
|
| 16 |
+
]
|