Spaces:
Paused
Paused
feat(sailorchat): using sailor chat model
Browse files- Dockerfile +1 -1
- run-sailor.sh +1 -1
Dockerfile
CHANGED
|
@@ -32,4 +32,4 @@ EXPOSE 7860
|
|
| 32 |
RUN chmod +x /app/run-llama.sh
|
| 33 |
RUN chmod +x /app/run-sailor.sh
|
| 34 |
|
| 35 |
-
CMD ["/app/run-
|
|
|
|
| 32 |
RUN chmod +x /app/run-llama.sh
|
| 33 |
RUN chmod +x /app/run-sailor.sh
|
| 34 |
|
| 35 |
+
CMD ["/app/run-sailor.sh"]
|
run-sailor.sh
CHANGED
|
@@ -27,4 +27,4 @@ python -u /app/openai_compatible_api_server.py \
|
|
| 27 |
--max-model-len 32768 \
|
| 28 |
--dtype float16 \
|
| 29 |
--enforce-eager \
|
| 30 |
-
--gpu-memory-utilization 0.
|
|
|
|
| 27 |
--max-model-len 32768 \
|
| 28 |
--dtype float16 \
|
| 29 |
--enforce-eager \
|
| 30 |
+
--gpu-memory-utilization 0.85
|