[Bug]: DeepSeek reasoning_content stream response returns None always

[MervinPraison]

What happened?

Here is the example code.

DeepSeek reasoning_content stream response returns None always

without streaming i can see the reasoning_content working fine!

from litellm import completion

resp = completion(
    model="deepseek/deepseek-reasoner",
    messages=[{"role": "user", "content": "What is 1+1?"}],
    stream=True
)

for chunk in resp:
    print(chunk)

Relevant log output

ModelResponseStream(id='xxxxx-36be-4fcf-xxxxx-4d36ecaf7e0b', created=2345235, model='deepseek-reasoner', object='chat.completion.chunk', system_fingerprint='fp_1c5d8833bc', choices=[StreamingChoices(finish_reason=None, index=0, delta=Delta(provider_specific_fields={'reasoning_content': None}, refusal=None, content='The', role='assistant', function_call=None, tool_calls=None, audio=None), logprobs=None)], stream_options=None, citations=None)

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

v1.59.5

Twitter / LinkedIn details

@MervinPraison

[MervinPraison]

Even the litellm.stream_chunk_builder function is not returning any reasoning_content

from litellm import completion
import litellm

messages = [{"role": "user", "content": "What is 1+1?"}]
resp = completion(
    model="deepseek/deepseek-reasoner",
    messages=messages,
    stream=True
)

chunks = []
for chunk in resp:
    chunks.append(chunk)

print(litellm.stream_chunk_builder(chunks, messages=messages))

[maykcaldas]

+1. If not streaming, the provider_specific_fields is populated successfully. But with stream, it does not

[MarcoWel]

+1
Our setup is OpenwebUI->Litellm->Ollama, it is close to unusable with Deepseek-R1.
OpenwebUI->Ollama works perfectly fine including live streaming of reasoning content.