Upload 2 files
Browse files- Dockerfile +12 -11
- config.py +20 -0
Dockerfile
CHANGED
@@ -1,12 +1,13 @@
|
|
1 |
-
FROM ghcr.io/sergey21000/gradio-llamacpp-chatbot:main-cpu
|
2 |
-
|
3 |
-
RUN useradd -m -u 1000 user \
|
4 |
-
&& chown -R user:user /app
|
5 |
-
|
6 |
-
USER user
|
7 |
-
ENV HOME=/home/user \
|
8 |
-
PATH=/home/user/.local/bin:$PATH
|
9 |
-
|
10 |
-
WORKDIR /app
|
11 |
-
|
|
|
12 |
CMD ["python3", "app.py"]
|
|
|
1 |
+
FROM ghcr.io/sergey21000/gradio-llamacpp-chatbot:main-cpu
|
2 |
+
|
3 |
+
RUN useradd -m -u 1000 user \
|
4 |
+
&& chown -R user:user /app
|
5 |
+
|
6 |
+
USER user
|
7 |
+
ENV HOME=/home/user \
|
8 |
+
PATH=/home/user/.local/bin:$PATH
|
9 |
+
|
10 |
+
WORKDIR /app
|
11 |
+
COPY --chown=user config.py ./
|
12 |
+
|
13 |
CMD ["python3", "app.py"]
|
config.py
ADDED
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
GENERATION_KWARGS = dict(
|
2 |
+
temperature=0.2,
|
3 |
+
top_p=0.95,
|
4 |
+
top_k=40,
|
5 |
+
repeat_penalty=1.0,
|
6 |
+
)
|
7 |
+
|
8 |
+
LLAMA_MODEL_KWARGS = dict(
|
9 |
+
# repo_id='bartowski/google_gemma-3-1b-it-GGUF',
|
10 |
+
# filename='google_gemma-3-1b-it-Q8_0.gguf',
|
11 |
+
repo_id='bartowski/Qwen_Qwen3-0.6B-GGUF',
|
12 |
+
filename='Qwen_Qwen3-0.6B-Q4_K_M.gguf',
|
13 |
+
local_dir='model',
|
14 |
+
cache_dir='model',
|
15 |
+
n_gpu_layers=-1,
|
16 |
+
verbose=True,
|
17 |
+
n_ctx=4096,
|
18 |
+
)
|
19 |
+
|
20 |
+
SHOW_THINKING = False
|