Transformers Local Client#

Source: examples/clients/transformers_local_client.py

Introduction#

Transformers pipelines are often the first local baseline for experimentation, HELM stresses the value of consistent evaluation scaffolding, and AI-assisted design education literature motivates reproducible local setups for pedagogy. This example demonstrates the Transformers local client path with deterministic trace output.

Technical Implementation#

  1. Configure Tracer with JSONL + console output so each run emits machine-readable traces and lifecycle logs.

  2. Build the runtime surface (public APIs only) and execute TransformersLocalLLMClient.generate(...) with a fixed request_id.

  3. Construct LLMRequest inputs and call generate through the selected client implementation.

  4. Print a compact JSON payload including trace_info for deterministic tests and docs examples.

        flowchart LR
    A["Input prompt or scenario"] --> B["main(): runtime wiring"]
    B --> C["TransformersLocalLLMClient.generate(...)"]
    C --> D["LLMRequest/LLMResponse contracts wrap provider behavior"]
    C --> E["Tracer JSONL + console events"]
    D --> F["ExecutionResult/payload"]
    E --> F
    F --> G["Printed JSON output"]
    
 1from __future__ import annotations
 2
 3import json
 4from pathlib import Path
 5
 6import design_research_agents as drag
 7
 8
 9def _build_payload() -> dict[str, object]:
10    # Run the local Transformers client using public runtime APIs. Using this with statement will automatically
11    # release any loaded model resources when the example is done.
12    with drag.TransformersLocalLLMClient(
13        name="transformers-local-dev",
14        model_id="Qwen/Qwen2.5-1.5B-Instruct",
15        default_model="Qwen/Qwen2.5-1.5B-Instruct",
16        device="auto",
17        dtype="auto",
18        quantization="none",
19        trust_remote_code=False,
20        revision="main",
21        max_retries=2,
22        model_patterns=("Qwen/*", "qwen2.5-*"),
23    ) as client:
24        description = client.describe()
25        prompt = "Provide one sentence on why deterministic local runs aid design reproducibility."
26        response = client.generate(
27            drag.LLMRequest(
28                messages=(
29                    drag.LLMMessage(role="system", content="You are a concise engineering design assistant."),
30                    drag.LLMMessage(role="user", content=prompt),
31                ),
32                model=client.default_model(),
33                temperature=0.0,
34                max_tokens=120,
35            )
36        )
37        llm_call = {
38            "prompt": prompt,
39            "response_text": response.text,
40            "response_model": response.model,
41            "response_provider": response.provider,
42            "response_has_text": bool(response.text.strip()),
43        }
44        return {
45            "client_class": description["client_class"],
46            "default_model": description["default_model"],
47            "llm_call": llm_call,
48            "backend": description["backend"],
49            "capabilities": description["capabilities"],
50            "server": description["server"],
51        }
52
53
54def main() -> None:
55    """Run traced Transformers client call payload."""
56    # Fixed request id keeps traces and docs output deterministic across runs.
57    request_id = "example-clients-transformers-local-call-001"
58    tracer = drag.Tracer(
59        enabled=True,
60        trace_dir=Path("artifacts/examples/traces"),
61        enable_jsonl=True,
62        enable_console=True,
63    )
64    payload = tracer.run_callable(
65        agent_name="ExamplesTransformersClientCall",
66        request_id=request_id,
67        input_payload={"scenario": "transformers-local-client-call"},
68        function=_build_payload,
69    )
70    assert isinstance(payload, dict)
71    payload["example"] = "clients/transformers_local_client.py"
72    payload["trace"] = tracer.trace_info(request_id)
73    # Print the results
74    print(json.dumps(payload, ensure_ascii=True, indent=2, sort_keys=True))
75
76
77if __name__ == "__main__":
78    main()

Expected Results#

Run Command

PYTHONPATH=src python3 examples/clients/transformers_local_client.py

Example output captured with DRA_EXAMPLE_LLM_MODE=deterministic (timestamps, durations, and trace filenames vary by run):

{
  "backend": {
    "base_url": null,
    "default_model": "Qwen/Qwen2.5-1.5B-Instruct",
    "device": "auto",
    "dtype": "auto",
    "kind": "transformers_local",
    "max_retries": 2,
    "model_id": "Qwen/Qwen2.5-1.5B-Instruct",
    "model_patterns": [
      "Qwen/*",
      "qwen2.5-*"
    ],
    "name": "transformers-local-dev",
    "quantization": "none"
  },
  "capabilities": {
    "json_mode": "prompt+validate",
    "max_context_tokens": null,
    "streaming": false,
    "tool_calling": "best_effort",
    "vision": false
  },
  "client_class": "TransformersLocalLLMClient",
  "default_model": "Qwen/Qwen2.5-1.5B-Instruct",
  "example": "clients/transformers_local_client.py",
  "llm_call": {
    "prompt": "Provide one sentence on why deterministic local runs aid design reproducibility.",
    "response_has_text": true,
    "response_model": "Qwen/Qwen2.5-1.5B-Instruct",
    "response_provider": "example-test-monkeypatch",
    "response_text": "Deterministic local runs make design comparisons repeatable across experiments."
  },
  "server": null,
  "trace": {
    "request_id": "example-clients-transformers-local-call-001",
    "trace_dir": "artifacts/examples/traces",
    "trace_path": "artifacts/examples/traces/run_20260222T162206Z_example-clients-transformers-local-call-001.jsonl"
  }
}

References#