Function get_bearer_token_provider is not JSON serializable

[arthurgreef]

A warning about using the AzureOpenAI is displayed in version 2.5.2. The recommendation is to use dspy.LM. I get the serialize exception when I change this the client from this:

from azure.identity import DefaultAzureCredential, get_bearer_token_provider

token_provider = get_bearer_token_provider(
    DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default"
)

model=os.environ["AZURE_AI_MODEL_GPT_4o"]

llm_gpt_4o = dspy.LM(
    model="gpt-4o"
    api_version=os.environ["OPENAI_API_VERSION"],
    api_base=os.environ["AZURE_OPENAI_ENDPOINT"],
    azure_ad_token_provider=token_provider,
    max_tokens=3000,
)

to this:

llm_gpt_4o = dspy.LM(
    model=os.environ["AZURE_AI_MODEL_GPT_4o"],
    api_version=os.environ["OPENAI_API_VERSION"],
    api_base=os.environ["AZURE_OPENAI_ENDPOINT"],
    azure_ad_token_provider=token_provider,
    max_tokens=3000,
)

[okhat]

Thanks @arthurgreef ! What is the serialize exception? Basically what's the error?

You can also get a much more complete error message by doing:

import litellm
litellm.set_verbose=True # 👈 this is the 1-line change you need to make

Also what's in os.environ["AZURE_AI_MODEL_GPT_4o"]? It should probably start with azure/ or something like that for LiteLLM.

[arthurgreef]

I took another look. This works great. Thanks.

from azure.identity import DefaultAzureCredential, get_bearer_token_provider

azure_ad_token = get_bearer_token_provider(
    DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default"
)()

model=os.environ["AZURE_AI_MODEL_GPT_4o"]

llm_gpt_4o = dspy.LM(
    model="azure/"+os.environ["AZURE_AI_MODEL_GPT_4o"],
    api_version=os.environ["OPENAI_API_VERSION"],
    api_base=os.environ["AZURE_OPENAI_ENDPOINT"],
    azure_ad_token=azure_ad_token,
    max_tokens=3000,
)

dspy.configure(lm=llm_gpt_4o)

[arthurgreef]

I am now having a problem with tracing in Phoenix using the adaptors. Is there something that needs to be updated in Pheonix. Here is the code that works with the AzureOpenAI client.

    # Needs to be before dspy import to get token tallies
    os.environ["DSP_CACHEBOOL"] = "False"
    import dspy

    from openinference.instrumentation.dspy import DSPyInstrumentor
    from openinference.instrumentation.openai import OpenAIInstrumentor
    from opentelemetry import trace as trace_api
    from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter
    from opentelemetry.sdk import trace as trace_sdk
    from opentelemetry.sdk.resources import Resource
    from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor

    from phoenix.otel import register
    endpoint="http://127.0.0.1:6006/v1/traces"
    register(endpoint=endpoint)

    resource = Resource(attributes={})
    tracer_provider = trace_sdk.TracerProvider(resource=resource)
    tracer_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter(endpoint)))
    tracer_provider.add_span_processor(SimpleSpanProcessor(ConsoleSpanExporter()))

    trace_api.set_tracer_provider(tracer_provider=tracer_provider)

    DSPyInstrumentor().instrument()
    OpenAIInstrumentor().instrument()

[okhat]

Yeah, the design of tracing (Phoenix and LangFuse) needs to change since the library is different internally for dspy.LM (DSPy 2.5 onwards).

By the way, you may also want to set:

os.environ["DSPY_CACHEBOOL"] = "False" # notice the Y, for 2.5 clients

[arthurgreef]

Requested a new feature to support DSPy 2.5.

[okhat]

Really appreciate your initiative on this @arthurgreef !

[arthurgreef]

This has been fixed https://github.com/Arize-ai/openinference/issues/1050