Update app.py

app.py

@@ -1,31 +1,24 @@
 import os
 import gradio as gr
-from huggingface_hub import InferenceClient
+from openai import OpenAI
 
-# In Space Settings → Secrets, add HF_TOKEN
 HF_TOKEN = os.getenv("HF_TOKEN", "")
+DEFAULT_MODEL = os.getenv("HF_MODEL", "Qwen/Qwen2.5-7B-Instruct")
 
-# You can also add HF_MODEL in Secrets; otherwise use a default.
-DEFAULT_MODEL = os.getenv("HF_MODEL", "Qwen/Qwen2.5-72B-Instruct")
-
-MAX_NEW_TOKENS = 1400
-TEMPERATURE = 0.4
+# Hugging Face Inference Providers router (OpenAI-compatible)
+# Docs: base_url="https://router.huggingface.co/v1"
+ROUTER_BASE_URL = "https://router.huggingface.co/v1"
 
 SYSTEM_INSTRUCTIONS = """
-You are a senior enterprise Account Executive specializing in converting
-open-source platform usage into enterprise-grade commercial engagements.
-
-You do NOT represent any specific vendor.
-You do NOT claim access to private or internal data.
+You are a senior enterprise Account Executive specializing in converting open-source platform usage into enterprise-grade commercial engagements.
 
-Hard rules:
-- Anchor everything in public signals
-- Separate technical users from decision makers
-- Be practical and executable
-- Avoid generic SaaS language
-- If information is missing, state assumptions clearly
+Rules:
+- Do NOT represent any specific vendor.
+- Do NOT claim access to private/internal data.
+- Anchor everything in public signals and state assumptions clearly.
+- Avoid generic SaaS buzzwords. Be executable.
 
-You MUST output exactly these sections:
+You MUST output exactly these headings:
 
 1. Account Snapshot
 2. Usage Hypotheses
@@ -54,52 +47,23 @@ Goal: {goal}
 Notes: {notes}
 
 Task:
-Create a structured outbound plan for this account based on public
-open-source platform usage and OSS → Enterprise conversion patterns.
+Create a structured outbound plan for this account based on public open-source platform usage and OSS → Enterprise conversion patterns.
 """.strip()
 
 
-def call_model(prompt: str, model: str) -> str:
-    # IMPORTANT: Don't crash the Space if the token is missing.
+def generate(account, industry, region, primary_persona, secondary_personas, offer_focus, goal, notes, model):
+    account = (account or "").strip()
+    if not account:
+        return "Please input an account name or org URL."
+
     if not HF_TOKEN:
         return (
             "Missing HF_TOKEN.\n\n"
             "Fix:\n"
-            "Space → Settings → Secrets → Add:\n"
-            "- Name: HF_TOKEN\n"
-            "- Value: your Hugging Face token\n"
+            "Space → Settings → Secrets → add HF_TOKEN with permission: “Make calls to Inference Providers”."
         )
 
-    client = InferenceClient(model=model, token=HF_TOKEN, base_url="https://router.huggingface.co")
-
-
-    # Most-compatible approach across endpoints:
-    full_prompt = f"{SYSTEM_INSTRUCTIONS}\n\n{prompt}\n\nReturn the output using the required 6 headings."
-
-    try:
-        out = client.text_generation(
-            full_prompt,
-            max_new_tokens=MAX_NEW_TOKENS,
-            temperature=TEMPERATURE,
-            do_sample=True,
-            return_full_text=False,
-        )
-        return (out or "").strip()
-    except Exception as e:
-        return (
-            "Model call failed.\n\n"
-            f"Model: {model}\n"
-            f"Error: {repr(e)}\n\n"
-            "Try:\n"
-            "1) Set a different HF_MODEL (an instruct model you have access to)\n"
-            "2) Or use a smaller model for reliability\n"
-        )
-
-
-def generate(account, industry, region, primary_persona, secondary_personas, offer_focus, goal, notes, model):
-    account = (account or "").strip()
-    if not account:
-        return "Please input an account name or org URL."
+    client = OpenAI(base_url=ROUTER_BASE_URL, api_key=HF_TOKEN)
 
     prompt = USER_TEMPLATE.format(
         account=account,
@@ -108,21 +72,41 @@ def generate(account, industry, region, primary_persona, secondary_personas, off
         primary_persona=primary_persona or "Not provided",
         secondary_personas=secondary_personas or "Not provided",
         offer_focus=offer_focus or "Not provided",
-        goal=goal or "Convert free usage to enterprise engagement",
+        goal=goal or "Convert free/community usage to enterprise engagement",
         notes=notes or "Not provided",
     )
 
     model = (model or DEFAULT_MODEL).strip()
-    return call_model(prompt, model)
+
+    try:
+        resp = client.chat.completions.create(
+            model=model,
+            messages=[
+                {"role": "system", "content": SYSTEM_INSTRUCTIONS},
+                {"role": "user", "content": prompt},
+            ],
+            temperature=0.4,
+            max_tokens=1400,
+        )
+        return resp.choices[0].message.content.strip()
+    except Exception as e:
+        return (
+            "Model call failed.\n\n"
+            f"Model: {model}\n"
+            f"Error: {repr(e)}\n\n"
+            "Try switching the model to a smaller instruct model you have access to, e.g.:\n"
+            "- Qwen/Qwen2.5-7B-Instruct\n"
+            "- mistralai/Mistral-7B-Instruct-v0.3\n"
+        )
 
 
 with gr.Blocks(title="Enterprise OSS → Outbound Ops") as app:
     gr.Markdown(
         "# Enterprise OSS → Outbound Ops\n"
-        "Input an **account name or public org URL** to generate an enterprise outbound plan."
+        "Input an **account name or org URL** to generate a structured enterprise outbound plan."
     )
 
-    account = gr.Textbox(label="Account name or org URL", placeholder="e.g., Emirates or https://huggingface.co/emirates")
+    account = gr.Textbox(label="Account name or org URL", placeholder="e.g., Emirates OR https://huggingface.co/emirates")
 
     with gr.Row():
         industry = gr.Textbox(label="Industry (optional)")
@@ -147,5 +131,4 @@ with gr.Blocks(title="Enterprise OSS → Outbound Ops") as app:
         outputs=output,
     )
 
-# IMPORTANT for Hugging Face Spaces:
 app.launch(server_name="0.0.0.0", server_port=7860)