init

Dockerfile

@@ -0,0 +1,25 @@
+# Use an alias for the base image for easier updates
+FROM python:3.10 as base
+
+# Set model
+ENV MODEL=macadeliccc/piccolo-2x7b-GGUF
+ENV QUANT=Q4_K_M
+ENV CHAT_TEMPLATE=chatml
+
+# Set the working directory
+WORKDIR /app
+
+# Install Python requirements
+COPY ./requirements.txt /app/
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+
+# Download model
+RUN MODEL_NAME_FILE=$(echo ${MODEL#*/} | tr '[:upper:]' '[:lower:]' | sed 's/-gguf$//') && \
+    wget https://huggingface.co/${MODEL}/resolve/main/${MODEL_NAME_FILE}.${QUANT}.gguf -O model.gguf
+
+# Copy the rest of your application
+COPY . .
+
+# Command to run the application
+CMD ["python", "app.py"]
+

app.py

@@ -0,0 +1,52 @@
+import os
+import json
+import gradio as gr
+from llama_cpp import Llama
+
+# Get environment variables
+model_id = os.getenv('MODEL')
+quant = os.getenv('QUANT')
+chat_template = os.getenv('CHAT_TEMPLATE')
+
+# Interface variables
+model_name = model_id.split('/')[1].split('-GGUF')[0]
+title = f"🔮 {model_name}"
+description = f"Chat with <a href=\"https://huggingface.co/{model_id}\">{model_name}</a> in GGUF format ({quant})!"
+
+# Initialize the LLM
+llm = Llama(model_path="model.gguf",
+            n_ctx=4096,
+            n_threads=2,
+            chat_format=chat_template)
+
+# Function for streaming chat completions
+def chat_stream_completion(message, history, system_prompt):
+    messages_prompts = [{"role": "system", "content": system_prompt}]
+    for human, assistant in history:
+        messages_prompts.append({"role": "user", "content": human})
+        messages_prompts.append({"role": "assistant", "content": assistant})
+    messages_prompts.append({"role": "user", "content": message})
+
+    response = llm.create_chat_completion(
+        messages=messages_prompts,
+        stream=True
+    )
+    message_repl = ""
+    for chunk in response:
+        if len(chunk['choices'][0]["delta"]) != 0 and "content" in chunk['choices'][0]["delta"]:
+            message_repl = message_repl + chunk['choices'][0]["delta"]["content"]
+        yield message_repl
+
+# Gradio chat interface
+gr.ChatInterface(
+    fn=chat_stream_completion,
+    title=title,
+    description=description,
+    additional_inputs=[gr.Textbox("You are helpful assistant.")],
+    additional_inputs_accordion="📝 System prompt",
+    examples=[
+        ["What is a Large Language Model?"],
+        ["What's 9+2-1?"],
+        ["Write Python code to print the Fibonacci sequence"]
+    ]
+).queue().launch(server_name="0.0.0.0")

requirements.txt

@@ -0,0 +1,2 @@
+llama-cpp-python
+gradio