Upload 4 files

Dockerfile

@@ -4,26 +4,33 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     python3 python3-pip curl ca-certificates \
     && rm -rf /var/lib/apt/lists/*
 
-# Download llama.cpp pre-built binary
+# Download llama.cpp pre-built binary (latest ubuntu x64)
 RUN curl -L https://github.com/ggml-org/llama.cpp/releases/download/b8468/llama-b8468-bin-ubuntu-x64.tar.gz \
     -o /tmp/llama.tar.gz \
     && mkdir -p /opt/llama \
-    && tar xzf /tmp/llama.tar.gz -C /opt/llama --strip-components=1 \
-    && rm /tmp/llama.tar.gz
+    && tar xzf /tmp/llama.tar.gz -C /opt/llama \
+    && rm /tmp/llama.tar.gz \
+    && find /opt/llama -name "llama-server" -exec dirname {} \; > /tmp/bindir.txt \
+    && echo "Binary dir: $(cat /tmp/bindir.txt)"
 
+# Make all llama binaries accessible
 ENV PATH="/opt/llama/build/bin:/opt/llama/bin:$PATH" \
-    LD_LIBRARY_PATH="/opt/llama/build/bin:/opt/llama/lib:$LD_LIBRARY_PATH"
+    LD_LIBRARY_PATH="/opt/llama/build/bin:/opt/llama/lib:/opt/llama:$LD_LIBRARY_PATH"
+
+# Add symlinks so llama-server is always found
+RUN find /opt/llama -name "llama-server" -exec ln -sf {} /usr/local/bin/llama-server \; \
+    && find /opt/llama -name "*.so*" -exec cp {} /usr/local/lib/ \; 2>/dev/null; ldconfig || true
 
 RUN useradd -m -u 1000 user
 USER user
-ENV HOME=/home/user PATH="/home/user/.local/bin:/opt/llama/build/bin:/opt/llama/bin:$PATH"
+ENV HOME=/home/user PATH="/home/user/.local/bin:/usr/local/bin:$PATH"
 WORKDIR /home/user/app
 
-RUN pip3 install --no-cache-dir --user \
-    gradio huggingface_hub openai
+RUN pip3 install --no-cache-dir --user gradio huggingface_hub openai
 
 COPY --chown=user app.py .
 COPY --chown=user start.sh .
+RUN chmod +x start.sh
 
 EXPOSE 7860
-CMD ["bash", "start.sh"]
+CMD ["bash", "start.sh"]

README.md

@@ -11,9 +11,4 @@ license: apache-2.0
 
 # Qwen3.5-9B Claude 4.6 Opus Reasoning Distilled v2
 
-A chatbot powered by **Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-v2** (Q4_K_M GGUF).
-
-- Claude 4.6 Opus reasoning patterns distilled into Qwen3.5-9B
-- Optimized for coding, math, and logical reasoning tasks
-- Supports Myanmar/Burmese language
-- Running on CPU with llama-cpp-python
+Powered by **Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-v2** (Q4_K_M GGUF) + llama.cpp server.

app.py

@@ -18,27 +18,50 @@ def chat(message, history, system_prompt, max_tokens, temperature, show_thinking
             messages.append({"role": "assistant", "content": h["content"]})
     messages.append({"role": "user", "content": message})
 
-    stream = client.chat.completions.create(
-        model="local",
-        messages=messages,
-        max_tokens=int(max_tokens),
-        temperature=float(temperature),
-        stream=True,
-    )
-
-    partial = ""
-    for chunk in stream:
-        delta = chunk.choices[0].delta.content or ""
-        partial += delta
-        yield strip_think_tags(partial) if not show_thinking else partial
+    try:
+        stream = client.chat.completions.create(
+            model="local",
+            messages=messages,
+            max_tokens=int(max_tokens),
+            temperature=float(temperature),
+            stream=True,
+        )
+        partial = ""
+        for chunk in stream:
+            delta = chunk.choices[0].delta.content or ""
+            partial += delta
+            if show_thinking:
+                yield partial
+            else:
+                yield strip_think_tags(partial)
+    except Exception as e:
+        yield f"Error: {str(e)} - llama-server is still loading, please wait and try again."
 
-with gr.Blocks(title="Qwen3.5-9B Claude Reasoning") as demo:
-    gr.Markdown("# Qwen3.5-9B Claude 4.6 Opus Reasoning v2\n> CPU inference - response 30s-2min ကြာနိုင်ပါတယ်")
-    system_prompt = gr.Textbox(label="System Prompt", value="You are a helpful coding and reasoning assistant. Respond in the same language the user uses.", lines=2)
+with gr.Blocks(title="Qwen3.5-9B Claude Reasoning", theme=gr.themes.Soft()) as demo:
+    gr.Markdown(
+        """
+        # Qwen3.5-9B Claude 4.6 Opus Reasoning Distilled v2
+        CPU inference ဖြစ်တဲ့အတွက် response 30s-2min ကြာနိုင်ပါတယ်။ မြန်မာလိုလည်း မေးလို့ရပါတယ်။
+        """
+    )
+    system_prompt = gr.Textbox(
+        label="System Prompt",
+        value="You are a helpful coding and reasoning assistant. Respond in the same language the user uses.",
+        lines=2,
+    )
     with gr.Row():
         max_tokens = gr.Slider(64, 2048, 512, step=64, label="Max Tokens")
         temperature = gr.Slider(0.1, 1.5, 0.7, step=0.1, label="Temperature")
         show_thinking = gr.Checkbox(label="Show thinking", value=False)
-    gr.ChatInterface(fn=chat, additional_inputs=[system_prompt, max_tokens, temperature, show_thinking])
 
-demo.launch(server_name="0.0.0.0", server_port=7860)
+    gr.ChatInterface(
+        fn=chat,
+        additional_inputs=[system_prompt, max_tokens, temperature, show_thinking],
+        examples=[
+            ["Python မှာ decorator ဘယ်လိုသုံးရလဲ"],
+            ["Write a Flask REST API with JWT auth"],
+            ["FizzBuzz ကို recursive function နဲ့ ရေးပြပါ"],
+        ],
+    )
+
+demo.launch(server_name="0.0.0.0", server_port=7860)

start.sh

@@ -1,28 +1,38 @@
 #!/bin/bash
+set -e
 
-echo "Downloading model..."
-python3 -c "
+echo "=== Downloading model ==="
+MODEL_PATH=$(python3 -c "
 from huggingface_hub import hf_hub_download
 path = hf_hub_download(
     repo_id='Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-v2-GGUF',
-    filename='Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-v2-Q4_K_M.gguf',
+    filename='Qwen3.5-9B.Q4_K_M.gguf',
 )
-print(f'MODEL_PATH={path}')
-" 2>&1 | tee /tmp/download.log
+print(path)
+")
 
-MODEL_PATH=$(grep "MODEL_PATH=" /tmp/download.log | tail -1 | cut -d= -f2)
+echo "Model path: $MODEL_PATH"
 
-echo "Starting llama-server with model: $MODEL_PATH"
+echo "=== Starting llama-server ==="
 llama-server \
     --model "$MODEL_PATH" \
     --host 127.0.0.1 \
     --port 8080 \
     --ctx-size 2048 \
     --threads 2 \
-    --batch-size 256 &
+    --batch-size 256 \
+    --log-disable &
 
-echo "Waiting for llama-server..."
-sleep 10
+# Wait for server to be ready
+echo "Waiting for llama-server to start..."
+for i in $(seq 1 30); do
+    if curl -s http://127.0.0.1:8080/health > /dev/null 2>&1; then
+        echo "llama-server is ready!"
+        break
+    fi
+    echo "  attempt $i/30..."
+    sleep 2
+done
 
-echo "Starting Gradio app..."
-python3 app.py
+echo "=== Starting Gradio app ==="
+exec python3 app.py