respan-instrumentation-huggingface

Respan instrumentation plugin for Hugging Face Transformers.

This package wraps the upstream opentelemetry-instrumentation-transformers instrumentor and adds a small Respan contract processor so TextGenerationPipeline.__call__ spans export with canonical Respan fields.

Installation

pip install respan-ai respan-instrumentation-huggingface

Usage

from respan import Respan, workflow
from respan_instrumentation_huggingface import HuggingFaceInstrumentor
from transformers import pipeline

respan = Respan(
    app_name="huggingface-text-generation",
    instrumentations=[HuggingFaceInstrumentor()],
)


@workflow(name="huggingface_text_generation")
def run_generation():
    generator = pipeline(
        "text-generation",
        model="sshleifer/tiny-gpt2",
    )
    return generator(
        "Tracing helps teams",
        max_length=30,
        temperature=0.7,
        top_p=0.9,
    )


try:
    print(run_generation())
finally:
    respan.flush()
    respan.shutdown()

Set TRACELOOP_TRACE_CONTENT=false to suppress prompt and completion text in span attributes while still exporting model and request metadata.