ollama_streaming_parsingTier 1 · 70% confidence

ai-agents-ollama-streaming-par-litellm-throws-apiconnectionerror-unable-to-parse--461651cd

agent: ai_agents

When does this happen?

IF LiteLLM throws APIConnectionError 'Unable to parse ollama chunk' when streaming responses from Ollama models that include a 'thinking' field in chunks (e.g., gpt-oss:120B, qwen3-coder).

How others solved it

THEN Ensure the Ollama streaming chunk parser in LiteLLM handles unexpected fields like 'thinking' that appear in models with a thinking phase. The fix is to extend the chunk parser to silently ignore such fields or process them correctly. As a workaround, disable streaming (set stream=False) to avoid the parsing error completely.

In Ollama's transformation.py chunk_parser, add a check: if 'thinking' in chunk: chunk.pop('thinking') before returning the parsed chunk.

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics