Skip to content

Instantly share code, notes, and snippets.

@Georgefwt
Created January 1, 2026 15:56
Show Gist options
  • Select an option

  • Save Georgefwt/fc397e8b14c8b0aa4357498ed63dcf27 to your computer and use it in GitHub Desktop.

Select an option

Save Georgefwt/fc397e8b14c8b0aa4357498ed63dcf27 to your computer and use it in GitHub Desktop.
Simple Sample Example using Tinker
import os
import tinker
from tinker import types
from tinker_cookbook import model_info, renderers
from tinker_cookbook.tokenizer_utils import get_tokenizer
# ========= Adjust here ===========
BASE_MODEL = "Qwen/Qwen3-235B-A22B-Instruct-2507"
MODEL_PATH = "SAVED SAMPLER CHECKPOINT PATH"
USER_QUERY = ""
# =================================
def _pick_message(parsed):
if isinstance(parsed, tuple):
return parsed[0]
if isinstance(parsed, list):
return parsed[0]
return parsed
def main():
if not os.getenv("TINKER_API_KEY"):
raise RuntimeError("Please set TINKER_API_KEY in your environment first.")
tokenizer = get_tokenizer(BASE_MODEL)
renderer_name = model_info.get_recommended_renderer_name(BASE_MODEL)
renderer = renderers.get_renderer(renderer_name, tokenizer)
convo = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": USER_QUERY},
]
model_input = renderer.build_generation_prompt(convo)
service_client = tinker.ServiceClient()
sampling_client = service_client.create_sampling_client(model_path=MODEL_PATH)
sampling_params = types.SamplingParams(
max_tokens=4096,
temperature=0.7,
top_p=0.9,
stop=renderer.get_stop_sequences(),
)
r = sampling_client.sample(prompt=model_input, num_samples=1, sampling_params=sampling_params).result()
seq = r.sequences[0]
parsed = renderer.parse_response(seq.tokens)
msg = _pick_message(parsed)
content = renderers.get_text_content(msg)
print("=== assistant ===")
print(content)
if __name__ == "__main__":
main()
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment