Upload 3 files

README.md

@@ -1,17 +1,24 @@
 ---
-title: Synthos Code Agent
-emoji: 💬
-colorFrom: yellow
-colorTo: purple
+title: Mistral Nemo Alpaca Finetune Demo
+emoji: 🦙
+colorFrom: purple
+colorTo: blue
 sdk: gradio
-sdk_version: 6.5.1
+sdk_version: 4.44.0
 app_file: app.py
 pinned: false
-hf_oauth: true
-hf_oauth_scopes:
-- inference-api
-license: apache-2.0
-short_description: A chatbot to generate code
+hardware: t4-medium
 ---
 
-An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).
+# Mistral Nemo Alpaca Finetune Demo
+
+This Space provides an interactive chat demo for the [TensorVizion/mistral-nemo-alpaca-finetune](https://huggingface.co/TensorVizion/mistral-nemo-alpaca-finetune) model.
+
+## Model Details
+- **Base Model:** Mistral Nemo (12B)
+- **Finetuned by:** TensorVizion
+- **Training:** Fine-tuned 2x faster with Unsloth and Hugging Face's TRL library on Alpaca-style instructions.
+- **Precision:** Loaded in 4-bit quantization (NF4) for efficient inference on consumer-grade GPUs.
+
+## Hardware Requirements
+This Space is configured to run on a `t4-medium` GPU (16GB VRAM). The 4-bit quantization ensures the 12B parameter model fits comfortably within this memory limit (~8GB VRAM usage). If you experience out-of-memory errors or want faster generation, consider upgrading to an `a10g-small` hardware tier in the Space settings.

app.py

@@ -1,69 +1,73 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
 
+MODEL_ID = "TensorVizion/mistral-nemo-alpaca-finetune"
 
-def respond(
-    message,
-    history: list[dict[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    hf_token: gr.OAuthToken,
-):
-    """
-    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-    """
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
-
-    messages = [{"role": "system", "content": system_message}]
-
-    messages.extend(history)
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
+# Configure 4-bit quantization to fit the 12B model in a T4 GPU (16GB VRAM)
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.bfloat16,
+    bnb_4bit_use_double_quant=True,
+)
 
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
+print("Loading tokenizer...")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
 
-        response += token
-        yield response
+print("Loading model in 4-bit...")
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    quantization_config=bnb_config,
+    device_map="auto",
+    torch_dtype=torch.bfloat16,
+)
 
+def generate_response(message, history):
+    # Format chat history for the model
+    chat_history = []
+    for user_msg, assistant_msg in history:
+        chat_history.append({"role": "user", "content": user_msg})
+        chat_history.append({"role": "assistant", "content": assistant_msg})
+    
+    chat_history.append({"role": "user", "content": message})
+    
+    # Apply the model's built-in chat template (handles Mistral Nemo formatting)
+    input_ids = tokenizer.apply_chat_template(
+        chat_history, 
+        add_generation_prompt=True, 
+        return_tensors="pt"
+    ).to(model.device)
+    
+    # Generate response
+    with torch.no_grad():
+        output_ids = model.generate(
+            input_ids,
+            max_new_tokens=512,
+            temperature=0.7,
+            top_p=0.9,
+            do_sample=True,
+            pad_token_id=tokenizer.eos_token_id
+        )
+    
+    # Decode only the newly generated tokens
+    response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
+    return response
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-chatbot = gr.ChatInterface(
-    respond,
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
+# Create Gradio Chat Interface
+demo = gr.ChatInterface(
+    fn=generate_response,
+    title="Mistral Nemo 12B Alpaca Finetune",
+    description="Chat with the TensorVizion Mistral Nemo 12B model. Runs efficiently in 4-bit quantization.",
+    examples=[
+        "Explain the concept of quantum entanglement in simple terms.",
+        "Write a short Python script to scrape a website's title.",
+        "What are the main differences between supervised and unsupervised learning?"
     ],
+    retry_btn=None,
+    undo_btn=None,
+    clear_btn="Clear Chat"
 )
 
-with gr.Blocks() as demo:
-    with gr.Sidebar():
-        gr.LoginButton()
-    chatbot.render()
-
-
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+torch>=2.4.0
+transformers>=4.44.0
+accelerate>=0.33.0
+bitsandbytes>=0.43.0
+gradio>=4.44.0