phase3_test_model.py

# ============================================================
# PHASE 3 — Testing the trained model
# Run: python phase3_test_model.py
# Make sure lora-adapter/ folder is in the same directory
# ============================================================

import torch
import json
import os
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

print("=" * 60)
print("  PHASE 3: Testing Prompt Optimizer")
print("=" * 60)

BASE_MODEL   = "google/gemma-2b"
ADAPTER_PATH = "./lora-adapter"

# ── Check adapter exists ────────────────────────────────────
if not os.path.exists(ADAPTER_PATH):
    print("\n❌ lora-adapter/ folder not found!")
    print("   → Download it from Colab after training finishes.")
    exit(1)

# ── Load model ──────────────────────────────────────────────
print("\n[1/3] Loading model + LoRA adapter (may take 1–2 min)...")
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
model     = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    torch_dtype=torch.float16,
    device_map="auto"
)
model = PeftModel.from_pretrained(model, ADAPTER_PATH)
model.eval()
print("      ✅ Model loaded")

# ── Inference function ──────────────────────────────────────
def improve_prompt(weak_prompt: str, max_new_tokens: int = 250) -> str:
    input_text = f"### Weak Prompt:\n{weak_prompt}\n\n### Improved Prompt:\n"
    inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
    with torch.no_grad():
        output = model.generate(
            **inputs,
            max_new_tokens=max_new_tokens,
            do_sample=True,
            temperature=0.7,
            top_p=0.9,
            repetition_penalty=1.1,
            pad_token_id=tokenizer.eos_token_id
        )
    full_text = tokenizer.decode(output[0], skip_special_tokens=True)
    # Return only the improved part
    if "### Improved Prompt:" in full_text:
        return full_text.split("### Improved Prompt:")[-1].strip()
    return full_text.strip()

# ── 15 test prompts (diverse topics) ───────────────────────
test_prompts = [
    "write about dogs",
    "explain machine learning",
    "help me code",
    "tell me about space",
    "make a diet plan",
    "write an email",
    "summarize history",
    "explain climate change",
    "how to learn python",
    "write a story",
    "explain blockchain",
    "give me recipe ideas",
    "help with my resume",
    "explain quantum computing",
    "plan a road trip",
]

print(f"\n[2/3] Testing on {len(test_prompts)} prompts...")
print("-" * 60)

results = []
for i, prompt in enumerate(test_prompts, 1):
    print(f"\n[{i}/{len(test_prompts)}] Weak: {prompt}")
    improved = improve_prompt(prompt)
    print(f"      Improved:\n{improved}")
    print("-" * 60)
    results.append({"weak": prompt, "improved": improved})

# ── Save results ────────────────────────────────────────────
os.makedirs("data", exist_ok=True)
with open("data/test_results.json", "w", encoding="utf-8") as f:
    json.dump(results, f, indent=2, ensure_ascii=False)

print("\n[3/3] Results saved")
print("\n" + "=" * 60)
print("  ✅ PHASE 3 COMPLETE!")
print("  📄 Results → data/test_results.json")
print("  ➡️  Next: Open dashboard.html to visualize results")
print("=" * 60)