fix: add thinking blocks #202
Open
fix: add thinking blocks #202
8000
+17
−5
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
There was a problem hiding this comment.
Pull request overview
This PR adds support for thinking_blocks from reasoning models (such as Claude Sonnet 4.5 and Claude Haiku 4.5) by properly extracting and preserving them in conversation history. This fixes a critical issue where reasoning models would fail due to missing thinking blocks in subsequent requests. Additionally, the PR corrects the message role for timeout notifications from "assistant" to "user".
- Adds
thinking_blocksfield toLLMResponsedataclass and extracts it from LLM responses for reasoning models - Updates
add_messagemethod to accept and storethinking_blocksin conversation history - Fixes message role from "assistant" to "user" for waiting timeout messages
Reviewed changes
Copilot reviewed 3 out of 3 changed files in this pull request and generated no comments.
| File | Description |
|---|---|
| strix/llm/llm.py | Adds thinking_blocks field to LLMResponse and extracts thinking blocks from reasoning model responses when applicable |
| strix/agents/state.py | Extends add_message to accept optional thinking_blocks parameter and conditionally includes them in message storage |
| strix/agents/base_agent.py | Passes thinking_blocks when storing assistant messages and corrects timeout message role from "assistant" to "user" |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR fixes a critical issue #156 that makes impossible to use important reasoning models, such as claude-4.5 or haiku, by returning
thinking_blockswhen the previous assistant response had them.Recently a related PR (BerriAI/litellm#17106) was merged in litellm. That PR basically removes thinking when thinking_blocks are not returned as a workaround to avoid errors. Having said that, that limits the capabilities of the reasoning LLMs. For that reason, litellm clients such as Strix need to handle the return of the thinking_blocks by themshelves, at least for reasoning models.
Ref: https://docs.litellm.ai/docs/reasoning_content
Ref: BerriAI/litellm#14194
Ref: BerriAI/litellm#9020
NOTE: I considered that one of the messages that was being sent as 'assistant' was incorrect and had to be sent as 'user' so I changed that as well. Please correct me if I'm wrong.
Tested with