Upload app.py with huggingface_hub

app.py

@@ -0,0 +1,188 @@
+"""
+HuggingFace Space - PineScript v5 Code Generator
+Gradio app for the fine-tuned model
+
+To deploy:
+1. Create a new Space on HuggingFace (Gradio SDK)
+2. Upload this file as app.py
+3. Add requirements.txt with: gradio, transformers, torch, accelerate, peft
+4. Set the model repo in the Space settings or as HF_MODEL_REPO secret
+"""
+
+import gradio as gr
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+from peft import AutoPeftModelForCausalLM
+import os
+
+# Configuration
+MODEL_REPO = "anthonym21/pinescript-v5-instructions-merged"
+USE_PEFT = False  # Merged model, no PEFT needed
+
+# Load model
+print(f"Loading model: {MODEL_REPO}")
+
+if torch.cuda.is_available():
+    # GPU available (paid Space or local)
+    bnb_config = BitsAndBytesConfig(
+        load_in_4bit=True,
+        bnb_4bit_quant_type="nf4",
+        bnb_4bit_compute_dtype=torch.bfloat16,
+    )
+
+    if USE_PEFT:
+        model = AutoPeftModelForCausalLM.from_pretrained(
+            MODEL_REPO,
+            quantization_config=bnb_config,
+            device_map="auto",
+            torch_dtype=torch.bfloat16,
+        )
+    else:
+        model = AutoModelForCausalLM.from_pretrained(
+            MODEL_REPO,
+            quantization_config=bnb_config,
+            device_map="auto",
+            torch_dtype=torch.bfloat16,
+        )
+else:
+    # CPU fallback (free Space - will be slow)
+    if USE_PEFT:
+        model = AutoPeftModelForCausalLM.from_pretrained(
+            MODEL_REPO,
+            device_map="cpu",
+            torch_dtype=torch.float32,
+        )
+    else:
+        model = AutoModelForCausalLM.from_pretrained(
+            MODEL_REPO,
+            device_map="cpu",
+            torch_dtype=torch.float32,
+        )
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO)
+tokenizer.pad_token = tokenizer.eos_token
+
+print("Model loaded!")
+
+
+def generate_pinescript(
+    prompt: str,
+    max_tokens: int = 1024,
+    temperature: float = 0.7,
+    top_p: float = 0.9,
+) -> str:
+    """Generate PineScript code from a prompt."""
+
+    # Format as instruction
+    formatted = f"""### Instruction:
+{prompt}
+
+### Response:
+"""
+
+    inputs = tokenizer(formatted, return_tensors="pt")
+    if torch.cuda.is_available():
+        inputs = inputs.to("cuda")
+
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=max_tokens,
+            temperature=temperature,
+            top_p=top_p,
+            do_sample=True,
+            pad_token_id=tokenizer.eos_token_id,
+            eos_token_id=tokenizer.eos_token_id,
+        )
+
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+
+    # Extract just the response part
+    if "### Response:" in response:
+        response = response.split("### Response:")[-1].strip()
+
+    return response
+
+
+# Example prompts
+EXAMPLES = [
+    ["Write a PineScript v5 indicator that shows RSI with overbought/oversold zones colored on the chart"],
+    ["Create a PineScript v5 strategy that buys when MACD crosses above signal and sells when it crosses below"],
+    ["Write a PineScript v5 indicator that displays Bollinger Bands with squeeze detection"],
+    ["Create a simple moving average crossover indicator in PineScript v5 with EMA 9 and EMA 21"],
+    ["Write a PineScript v5 indicator that shows support and resistance levels based on pivot points"],
+]
+
+# Gradio interface
+with gr.Blocks(title="PineScript v5 Generator", theme=gr.themes.Soft()) as demo:
+    gr.Markdown("""
+    # 🌲 PineScript v5 Code Generator
+
+    Generate TradingView PineScript v5 code using a fine-tuned CodeGemma model.
+
+    **Tips:**
+    - Be specific about what you want (indicator, strategy, specific features)
+    - Mention inputs, colors, and plot styles if you have preferences
+    - Ask for alerts, labels, or tables if needed
+    """)
+
+    with gr.Row():
+        with gr.Column(scale=2):
+            prompt = gr.Textbox(
+                label="What do you want to create?",
+                placeholder="e.g., Write a PineScript v5 indicator that shows RSI with dynamic overbought/oversold levels",
+                lines=3,
+            )
+
+            with gr.Row():
+                max_tokens = gr.Slider(
+                    minimum=256,
+                    maximum=2048,
+                    value=1024,
+                    step=128,
+                    label="Max Tokens",
+                )
+                temperature = gr.Slider(
+                    minimum=0.1,
+                    maximum=1.5,
+                    value=0.7,
+                    step=0.1,
+                    label="Temperature",
+                )
+                top_p = gr.Slider(
+                    minimum=0.1,
+                    maximum=1.0,
+                    value=0.9,
+                    step=0.05,
+                    label="Top P",
+                )
+
+            generate_btn = gr.Button("Generate PineScript", variant="primary")
+
+        with gr.Column(scale=3):
+            output = gr.Code(
+                label="Generated PineScript v5 Code",
+                language="javascript",  # Closest to PineScript syntax
+                lines=25,
+            )
+
+    gr.Examples(
+        examples=EXAMPLES,
+        inputs=[prompt],
+        label="Example Prompts",
+    )
+
+    generate_btn.click(
+        fn=generate_pinescript,
+        inputs=[prompt, max_tokens, temperature, top_p],
+        outputs=output,
+    )
+
+    gr.Markdown("""
+    ---
+    **Note:** This model was fine-tuned on the [PineScripts-Permissive](https://huggingface.co/datasets/mrmegatelo/PineScripts-Permissive) dataset.
+    Always review and test generated code before using in live trading.
+    """)
+
+if __name__ == "__main__":
+    demo.launch()