Create app1.py

app1.py

@@ -0,0 +1,171 @@
+# app.py
+import gradio as gr
+from transformers import AutoTokenizer, AutoConfig, AutoModelForCausalLM, pipeline
+from peft import PeftModel
+import torch
+import os
+import re
+import json
+import time
+from datetime import datetime
+from huggingface_hub import hf_hub_download, model_info
+
+# ====== Settings ======
+device = 0 if torch.cuda.is_available() else -1
+base_model_name = "Qwen/Qwen2.5-0.5B"  # Base model for LoRA
+finetuned_repo = "rahul7star/Qwen2.5-3B-Gita"
+
+log_lines = []
+
+def log(msg):
+    """Append timestamped message to log."""
+    line = f"[{datetime.now().strftime('%H:%M:%S')}] {msg}"
+    print(line)
+    log_lines.append(line)
+
+# ====== Start Logging ======
+log("🔍 Initializing model load sequence...")
+log(f"Base model: {base_model_name}")
+log(f"Fine-tuned LoRA repo: {finetuned_repo}")
+log(f"Device detected: {'GPU' if device==0 else 'CPU'}")
+hf_cache = os.path.expanduser("~/.cache/huggingface/hub")
+log(f"Model cache directory: {hf_cache}")
+
+# ====== Inspect Hugging Face repo ======
+try:
+    info = model_info(finetuned_repo)
+    log("📦 Hugging Face repo info loaded:")
+    log(f"  - Model ID: {info.id}")
+    log(f"  - Private: {info.private}")
+    log(f"  - Last modified: {info.last_modified}")
+    log(f"  - Files count: {len(info.siblings)}")
+    for s in info.siblings[:5]:
+        log(f"    · {s.rfilename}")
+except Exception as e:
+    log(f"⚠️ Could not fetch model info: {e}")
+
+# ====== Load base model and tokenizer ======
+try:
+    tokenizer = AutoTokenizer.from_pretrained(base_model_name, trust_remote_code=True)
+    if tokenizer.pad_token is None:
+        tokenizer.pad_token = tokenizer.eos_token
+    log(f"✅ Tokenizer loaded: vocab size {tokenizer.vocab_size}")
+except Exception as e:
+    log(f"❌ Failed to load tokenizer: {e}")
+    tokenizer = None
+
+try:
+    base_model = AutoModelForCausalLM.from_pretrained(
+        base_model_name,
+        trust_remote_code=True,
+        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+        device_map="auto" if torch.cuda.is_available() else None,
+    )
+    log(f"✅ Base model loaded from {base_model_name}")
+except Exception as e:
+    log(f"❌ Failed to load base model: {e}")
+    base_model = None
+
+# ====== Load fine-tuned LoRA weights ======
+model = None
+pipe = None
+try:
+    if base_model is not None:
+        model = PeftModel.from_pretrained(base_model, finetuned_repo)
+        model.eval()
+        log(f"✅ LoRA fine-tuned model loaded from {finetuned_repo}")
+        log(f"🧩 Model architecture: {getattr(model.config, 'architectures', ['Unknown'])}")
+
+        pipe = pipeline(
+            "text-generation",
+            model=model,
+            tokenizer=tokenizer,
+            device=device,
+        )
+        log("✅ Pipeline ready for inference")
+except Exception as e:
+    log(f"❌ Failed to load LoRA model: {e}")
+
+# ====== Try to extract training info ======
+def extract_training_info(repo_name):
+    data = {}
+    try:
+        readme_path = hf_hub_download(repo_name, filename="README.md")
+        with open(readme_path, "r", encoding="utf-8") as f:
+            text = f.read()
+            matches = re.findall(r"(rahul7star/\w+|dataset|fine[- ]?tune|trained on|data:)", text, re.I)
+            if matches:
+                data["readme_mentions"] = matches[:5]
+                log(f"✅ README mentions dataset/fine-tune: {matches[:5]}")
+            else:
+                log("ℹ️ No dataset reference found in README")
+    except Exception as e:
+        log(f"⚠️ README not found or unreadable: {e}")
+    return data
+
+training_info = extract_training_info(finetuned_repo)
+
+# ====== Chat Function ======
+def chat_with_model(message, history):
+    log_lines.clear()
+    log("💭 Starting chat generation...")
+    log(f"User message: {message}")
+
+    if pipe is None:
+        return "", history, "⚠️ Model pipeline not loaded. Check logs."
+
+    # Build context
+    context = "The following is a conversation between a user and an AI assistant inspired by the Bhagavad Gita.\n"
+    for user, bot in history:
+        context += f"User: {user}\nAssistant: {bot}\n"
+    context += f"User: {message}\nAssistant:"
+    log("📄 Context built:")
+    log(context)
+
+    # Generate
+    log("🧠 Generating response...")
+    start_time = time.time()
+    try:
+        output = pipe(
+            context,
+            max_new_tokens=200,
+            do_sample=True,
+            temperature=0.7,
+            top_p=0.9,
+            repetition_penalty=1.1,
+        )[0]["generated_text"]
+        log(f"⏱️ Inference took {time.time() - start_time:.2f}s")
+    except Exception as e:
+        log(f"❌ Generation failed: {e}")
+        return "", history, "\n".join(log_lines)
+
+    # Clean output
+    reply = output[len(context):].strip()
+    reply = re.sub(r"(ContentLoaded|<\/?[^>]+>|[\r\n]{2,})", " ", reply)
+    reply = re.sub(r"\s{2,}", " ", reply).strip()
+    reply = reply.split("User:")[0].split("Assistant:")[0].strip()
+
+    log("🪄 Output cleaned successfully")
+    log(f"Model reply: {reply}")
+
+    history.append((message, reply))
+    return "", history, "\n".join(log_lines)
+
+# ====== Gradio Interface ======
+with gr.Blocks(theme=gr.themes.Soft(primary_hue="blue")) as demo:
+    gr.Markdown("## 💬 Qwen Gita — LoRA Fine-tuned Conversational Assistant")
+
+    with gr.Row():
+        with gr.Column(scale=2):
+            chatbot = gr.Chatbot(height=500)
+            msg = gr.Textbox(placeholder="Ask about the Gita, life, or philosophy...", label="Your Message")
+            clear = gr.Button("Clear")
+        with gr.Column(scale=1):
+            log_box = gr.Textbox(label="Detailed Model Log", lines=25, interactive=False)
+
+    msg.submit(chat_with_model, [msg, chatbot], [msg, chatbot, log_box])
+    clear.click(lambda: (None, None, ""), None, [chatbot, log_box], queue=False)
+
+# ====== Launch ======
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)