# hf_diag.py
from huggingface_hub import InferenceClient
import os, json

TOK = os.getenv("HF_TOKEN") or "paste-your-token-here"
MODEL = "Qwen/Qwen2.5-Coder-32B-Instruct"   # try both your primary and fallback model names
client = InferenceClient(token=TOK)

def try_chat():
    messages = [{"role":"system","content":"You are a helpful assistant."},
                {"role":"user","content":"Say hello and return two short code files in JSON: files mapping with main.py and tests/test_main.py"}]
    try:
        print("Calling chat_completion (non-stream)...")
        resp = client.chat_completion(messages=messages, model=MODEL, stream=False, max_new_tokens=256)
        print("TYPE:", type(resp))
        try:
            print("As repr:", repr(resp)[:1000])
        except:
            print("Couldn't repr resp")
        # try to JSON-dump
        try:
            print("JSON-able?", json.dumps(resp)[:1000])
        except Exception as e:
            print("Not JSON serializable:", e)
    except Exception as e:
        print("chat_completion failed:", e)

if __name__ == "__main__":
    try_chat()