Add chat streaming for completions - #471
Closed
elimoss wants to merge 21 commits into
Closed
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
Adds a new streaming API to LLMModel/LiteLLMModel that yields incremental text deltas from provider streaming, followed by a single canonical terminal LLMResult containing the completed Message/ToolRequestMessage (keeping existing non-streaming call paths intact).
Changes:
- Introduces
LLMModel.call_stream()and plumbing to request Chat Completions streaming with delta-yielding support. - Refactors Chat Completion parsing into
_parse_chat_completion()and enhances custom tool parser type detection. - Improves stream lifecycle handling (
aclose) and adds/updates tests + README docs for streaming behavior.
Reviewed changes
Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.
Show a summary per file
| File | Description |
|---|---|
| packages/lmi/src/lmi/llms.py | Adds call_stream, stream config validation, stream delta yielding in acompletion_iter, stream closure improvements, and parsing refactor. |
| packages/lmi/src/lmi/cost_tracker.py | Adds TrackedStreamWrapper.aclose() to support contextlib.aclosing on tracked provider streams. |
| packages/lmi/tests/test_llms.py | Adds realistic stream-chunk builders and comprehensive call_stream tests (text + tool call assembly + closure + validation/callback behavior). |
| packages/lmi/tests/test_dispatch.py | Adds coverage ensuring _commit_stream closes the underlying source iterator when the committed stream is closed early. |
| packages/lmi/tests/test_cost_tracking.py | Adds coverage ensuring TrackedStreamWrapper.aclose() forwards to the wrapped stream. |
| packages/lmi/README.md | Fixes a typo and documents call_stream usage/semantics. |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
sidnarayanan
approved these changes
Jul 19, 2026
sidnarayanan
approved these changes
Jul 24, 2026
sidnarayanan
left a comment
sidnarayanan
left a comment
Collaborator
There was a problem hiding this comment.
Approving to unblock, but can you do a pass on documentation?
Require list[Message] on call_stream, matching call; the str convenience stays on call_single. Stop setting seconds_to_first_token and seconds_to_last_token in the call_stream path, since nothing reads them. Explain why consumers must close the iterator early, and restore comments and formatting that the streaming rewrite disturbed.
The streaming path skipped messages with no content, which drops a message that holds only tool calls. The tool responses answering it are then left without a request and providers reject the conversation. The non-streaming path has always sent every message, so this makes the two match.
Collaborator
Author
|
got through a prototype and it just wasn't really that much faster in practice. |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
yield incremental completion tokens as they stream in from the provider followed by a
LLMResultcontaining the completedMessageorToolRequestMessage. The former supports streaming UIs while the latter constitutes the canonical completion. existing calls in non-streaming contexts are unchanged.also, since we are handling streams now, changed some types around iterators to require aclose so we can cleanly close provider streams when necessary. this is achieved with a
ClosableAsyncIteratorthat narrowly requiresaclose, andAsyncGeneratorin cases where that is the actual return type