Update app.py

app.py

@@ -1,134 +1,93 @@
 import gradio as gr
 import os
-import requests
-
-HF_TOKEN = os.environ.get("HF_TOKEN", "")
-
-# Using a reliable free model on HuggingFace
-API_URL = "https://api-inference.huggingface.co/models/HuggingFaceH4/zephyr-7b-beta"
-
-def query_model(prompt):
-    if not HF_TOKEN:
-        return None, (
-            "⚠️ **HF_TOKEN is not set.**\n\n"
-            "Space Settings → Secrets → New secret\n"
-            "- Name: `HF_TOKEN`\n"
-            "- Value: your token from huggingface.co/settings/tokens"
-        )
-
-    headers = {"Authorization": f"Bearer {HF_TOKEN}"}
-
-    full_prompt = f"<|system|>\nYou are an expert API architect and startup advisor.\n</s>\n<|user|>\n{prompt}\n</s>\n<|assistant|>\n"
-
-    payload = {
-        "inputs": full_prompt,
-        "parameters": {
-            "max_new_tokens": 1000,
-            "temperature": 0.7,
-            "do_sample": True,
-            "return_full_text": False,
-            "stop": ["<|user|>", "</s>"]
-        }
-    }
-
-    try:
-        response = requests.post(API_URL, headers=headers, json=payload, timeout=60)
-
-        if response.status_code == 503:
-            return None, "⏳ Model is loading (cold start). Wait 30 seconds and try again."
-        if response.status_code == 401:
-            return None, "❌ Invalid HF_TOKEN. Go to Space Settings and check your token."
-        if response.status_code == 429:
-            return None, "❌ Rate limit. Wait 1 minute and try again."
-        if response.status_code != 200:
-            return None, f"❌ API Error {response.status_code}: {response.text[:300]}"
-
-        data = response.json()
-
-        if isinstance(data, list) and len(data) > 0:
-            text = data[0].get("generated_text", "").strip()
-            if text:
-                return text, None
-
-        if isinstance(data, dict) and "error" in data:
-            return None, f"❌ Model error: {data['error']}"
-
-        return None, f"❌ Empty response from model. Try again."
-
-    except requests.exceptions.Timeout:
-        return None, "⏳ Request timed out. The model may be loading. Try again in 30 seconds."
-    except Exception as e:
-        return None, f"❌ Error: {str(e)}"
+from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
+import torch
+
+MODEL_ID = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
+
+print("Loading model...")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+pipe = pipeline(
+    "text-generation",
+    model=MODEL_ID,
+    torch_dtype=torch.float32,
+    device_map="auto"
+)
+print("Model loaded.")
+
+SYSTEM = """You are an expert API architect. When given a product idea, recommend the best API stack with pricing and build order."""
+
+def generate(prompt):
+    messages = [
+        {"role": "system", "content": SYSTEM},
+        {"role": "user", "content": prompt}
+    ]
+    formatted = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+    out = pipe(
+        formatted,
+        max_new_tokens=800,
+        do_sample=True,
+        temperature=0.7,
+        return_full_text=False
+    )
+    return out[0]["generated_text"].strip()
 
 
 def get_stack(idea, budget, level, market):
     if not idea.strip():
         return "⚠️ Please describe your product idea."
+    try:
+        prompt = f"""Product idea: {idea}
+Budget: {budget} | Level: {level} | Market: {market}
 
-    prompt = f"""I want to build: {idea}
-
-My monthly API budget: {budget}
-My technical level: {level}
-My target market: {market}
-
-Please recommend the best API stack for my product using this format:
-
-## 🎯 My Idea
-(one sentence summary)
+Recommend 3-5 APIs using this format:
 
-## 🔧 Recommended APIs
+## 🎯 Idea Summary
+(one line)
 
-### [API Name] — [what it does in my product]
-- What it does: (one sentence)
-- Pricing: (free tier details + paid pricing)
+## 🔧 API Stack
+### [API Name] — [role]
+- Pricing: (free tier + paid)
 - Docs: (URL)
-- Difficulty: Easy / Medium / Hard
-
-(repeat for 3-5 APIs)
+- Difficulty: Easy/Medium/Hard
 
-## 💰 Monthly Cost Estimate
+## 💰 Cost Estimate
 - 100 users: $X/month
 - 1,000 users: $X/month
-- 10,000 users: $X/month
 
 ## ⚡ Build Order
-1. Start with [API] because...
-2. Then add [API] because...
-3. Finally [API] because...
+1. First: [API]
+2. Second: [API]
 
-## 🚨 Biggest Mistake to Avoid
+## 🚨 Top Mistake
 (one paragraph)"""
-
-    result, error = query_model(prompt)
-    if error:
-        return error
-    return result if result else "❌ No response received. Please try again."
+        return generate(prompt)
+    except Exception as e:
+        return f"❌ Error: {str(e)}"
 
 
 def ask_followup(question, prev):
     if not question.strip():
         return ""
     if not prev or prev.startswith("⚠️") or prev.startswith("❌"):
-        return "⚠️ Generate a stack recommendation first, then ask follow-up questions."
-
-    prompt = f"""Based on this API stack recommendation:
-{prev[:600]}
-
-Answer this follow-up question clearly and concisely:
-{question}"""
-
-    result, error = query_model(prompt)
-    if error:
-        return error
-    return result if result else "❌ No response. Please try again."
+        return "⚠️ Generate a recommendation first."
+    try:
+        prompt = f"Previous recommendation:\n{prev[:500]}\n\nFollow-up: {question}\n\nAnswer concisely."
+        return generate(prompt)
+    except Exception as e:
+        return f"❌ Error: {str(e)}"
 
 
 EXAMPLES = [
-    ["A Notion-like notes app with AI summarization", "$50/month", "Intermediate", "Students"],
-    ["SMS marketing platform for small restaurants", "$30/month", "Beginner", "Restaurant owners"],
-    ["Job board that auto-matches candidates with AI", "$100/month", "Advanced", "Tech recruiters"],
-    ["Podcast transcription and highlight tool", "$20/month", "Beginner", "Podcasters"],
-    ["Crypto portfolio tracker with price alerts", "$0 (free only)", "Intermediate", "Retail investors"],
+    ["A notes app with AI summarization", "$50/month", "Intermediate", "Students"],
+    ["SMS marketing for restaurants", "$30/month", "Beginner", "Restaurant owners"],
+    ["AI job matching platform", "$100/month", "Advanced", "Tech recruiters"],
+    ["Podcast transcription tool", "$20/month", "Beginner", "Podcasters"],
+    ["Crypto portfolio tracker", "$0 (free only)", "Intermediate", "Retail investors"],
 ]
 
 css = """
@@ -139,24 +98,22 @@ footer { display: none !important; }
 with gr.Blocks(css=css, title="API Stack Finder") as demo:
     gr.Markdown("""
 # 🔧 API Stack Finder
-**Describe your product idea → Get the perfect API stack, pricing, and build order**
+**Describe your product idea → Get the perfect API stack**
 
-Powered by **Zephyr-7B** via HuggingFace Inference API — 100% Free.
+Powered by **TinyLlama** — runs locally inside this Space, no external API needed.
 """)
 
     with gr.Row():
         with gr.Column(scale=3):
             idea_input = gr.Textbox(
                 label="Describe your product idea",
-                placeholder="e.g. A platform where freelancers sell services and get paid instantly...",
+                placeholder="e.g. A freelancer marketplace with instant payments...",
                 lines=4
             )
         with gr.Column(scale=1):
             budget_input = gr.Dropdown(
-                choices=[
-                    "$0 (free only)", "$10/month", "$30/month",
-                    "$50/month", "$100/month", "$500/month", "Unlimited"
-                ],
+                choices=["$0 (free only)", "$10/month", "$30/month",
+                         "$50/month", "$100/month", "$500/month", "Unlimited"],
                 value="$30/month",
                 label="Monthly API budget"
             )
@@ -174,7 +131,7 @@ Powered by **Zephyr-7B** via HuggingFace Inference API — 100% Free.
     gr.Markdown("**Try an example:**")
     with gr.Row():
         for ex in EXAMPLES:
-            gr.Button(ex[0][:32] + "...", size="sm").click(
+            gr.Button(ex[0][:30] + "...", size="sm").click(
                 lambda e=ex: (e[0], e[1], e[2], e[3]),
                 outputs=[idea_input, budget_input, level_input, market_input]
             )
@@ -184,7 +141,6 @@ Powered by **Zephyr-7B** via HuggingFace Inference API — 100% Free.
 
     gr.Markdown("---")
     gr.Markdown("### 💬 Ask a follow-up")
-
     with gr.Row():
         followup_input = gr.Textbox(
             label="Follow-up question",
@@ -192,25 +148,16 @@ Powered by **Zephyr-7B** via HuggingFace Inference API — 100% Free.
             scale=4
         )
         followup_btn = gr.Button("Ask", scale=1, variant="secondary")
-
     followup_output = gr.Markdown()
 
-    submit_btn.click(
-        get_stack,
+    submit_btn.click(get_stack,
         inputs=[idea_input, budget_input, level_input, market_input],
-        outputs=output
-    )
-    followup_btn.click(
-        ask_followup,
+        outputs=output)
+    followup_btn.click(ask_followup,
         inputs=[followup_input, output],
-        outputs=followup_output
-    )
+        outputs=followup_output)
 
-    gr.Markdown("""
----
-🤖 Model: `Zephyr-7B-Beta` · 🆓 Free via HuggingFace  
-🔑 Get token: [huggingface.co/settings/tokens](https://huggingface.co/settings/tokens)
-""")
+    gr.Markdown("---\n🤖 Model: `TinyLlama-1.1B-Chat` · Runs fully inside HuggingFace Space")
 
 if __name__ == "__main__":
     demo.launch()