space_client.py

"""
Call the deployed ZeroGPU Space API from localhost.

  pip install gradio_client
  export HF_TOKEN=hf_xxx     # your Pro token -> uses YOUR ZeroGPU quota

  python space_client.py
"""

import os

from gradio_client import Client, handle_file

SPACE = os.environ.get("SPACE_ID", "chabab/qwen-agentworld-35b-zerogpu")

# Passing hf_token charges the call to your account's ZeroGPU quota (recommended;
# anonymous calls share a much smaller pool and may be rejected).
client = Client(SPACE, token=os.environ.get("HF_TOKEN"))

# --- text only -------------------------------------------------------------
result = client.predict(
    message={"text": "Give me a one-sentence hello.", "files": []},
    max_new_tokens=256,
    temperature=0.7,
    api_name="/chat",
)
print("TEXT ->", result)

# --- image + text ----------------------------------------------------------
# result = client.predict(
#     message={
#         "text": "Describe this image.",
#         "files": [handle_file("https://gradio-builds.s3.amazonaws.com/demo-files/cheetah.jpg")],
#     },
#     max_new_tokens=256,
#     temperature=0.7,
#     api_name="/chat",
# )
# print("VISION ->", result)