Update app.py

app.py

@@ -1,25 +1,39 @@
+import os
 import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, BitsAndBytesConfig
-from peft import PeftModel
 import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+from peft import PeftModel
 
-base_model = "mistralai/Mistral-7B-v0.1"
-adapter_path = "./mistral-recovery-model"
+# 🔐 Hugging Face token (stored in repo secret as HF_TOKEN)
+HF_TOKEN = os.environ.get("HF_TOKEN")
 
-bnb_config = BitsAndBytesConfig(
-    load_in_4bit=True,
-    bnb_4bit_compute_dtype=torch.float16,
-)
+# 📌 Model sources
+base_model = "mistralai/Mistral-7B-v0.1"
+adapter_path = "./mistral-recovery-model"  # Your uploaded LoRA adapter files
 
-tokenizer = AutoTokenizer.from_pretrained(base_model)
+# 🚀 Load tokenizer and model in FP32 (CPU-safe)
+tokenizer = AutoTokenizer.from_pretrained(base_model, use_auth_token=HF_TOKEN)
 tokenizer.pad_token = tokenizer.eos_token
 
-model = AutoModelForCausalLM.from_pretrained(base_model, device_map="auto", quantization_config=bnb_config)
-model = PeftModel.from_pretrained(model, adapter_path)
+model = AutoModelForCausalLM.from_pretrained(
+    base_model,
+    torch_dtype=torch.float32,
+    use_auth_token=HF_TOKEN
+)
+
+model = PeftModel.from_pretrained(model, adapter_path, use_auth_token=HF_TOKEN)
 
+# 🧠 Pipeline
 pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, max_new_tokens=150, temperature=0.7)
 
+# 🖥️ Gradio UI
 def generate(prompt):
     return pipe(prompt)[0]['generated_text']
 
-gr.Interface(fn=generate, inputs="text", outputs="text", title="🏋️‍♂️ Mistral Recovery Coach").launch()
+gr.Interface(
+    fn=generate,
+    inputs="text",
+    outputs="text",
+    title="🏋️ Mistral Recovery Coach (CPU Mode)",
+    description="Enter your workout summary and get personalized recovery advice."
+).launch()