The AWS event stream framer now yields aws_smithy_types Message directly
instead of copying headers and payload into a second struct, which halves
its allocations (16 to 8 per frame) and cuts wall time by about 30%. It
also parses from a slice and advances the buffer rather than splitting and
freezing it.
SSE parsing no longer goes through sse-stream. The new parser follows the
WHATWG event stream grammar (unknown fields ignored, colonless lines are
empty fields, last event line wins, non-digit retry and NUL ids ignored,
events without data are not dispatched), caps an event at MAX_FRAME_BYTES
so an unterminated upstream cannot grow memory without bound, and reports
upstream body failures as Error::Body like the AWS path does. Allocations
stay at two per event and wall time is within a few percent.
The Framer trait had no generic consumer, so both parsers are plain
functions now. MissingStreamData in litellm-core is gone because a
dispatched SSE event always carries data.
The Rust bridge imported httpx to construct the provider error response, which
fails in the isolated wheel check where httpx is absent. Rust now raises
RustUpstreamError with a headers attribute and the Python lifecycle wraps it in
a typed UpstreamFailure carrying the httpx.Response before legacy mapping.
Test helpers gained call_native so pytest.raises blocks hold a single call
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Delete litellm/ocr/input.py and the native _ocr_file_document, _ocr_upload_document
and _ocr_mime_type helpers. File documents now project to a typed OcrDocumentInput
and the core lifecycle reads local paths, encodes bytes and asks the host to read
file-like objects through a ReadDocument operation before the provider request
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Move each route's bridge module under litellm/rust_bridge/<route>/ so a folder
means a Rust implementation exists while the catalog row says whether it is
used. OCR now keeps the Python implementation in litellm/ocr/main.py and the
Rust selection in litellm/ocr/rust.py, removing litellm/ocr/legacy.py
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Split the Rust workflow into fmt, clippy, nextest and wheel jobs so they run in parallel, replace manual actions/cache with Swatinem/rust-cache, and install a pinned checksum-verified cargo-nextest. Make two python-bridge tests self-contained so they pass when nextest runs each test in its own process.
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* feat(rust): count tiktoken cl100k_base admission tokens in Rust
The Rust admission token counter only had the Anthropic tokenizer, so every
other model (OpenAI gpt-4 family, Azure, Gemini, Bedrock non-Claude, Mistral)
tokenized with tiktoken on the Python inference worker.
Add an exact cl100k_base counter to litellm-token-counter: the vendored rank
file (base64 token / rank lines, the bytes Python's tiktoken uses) is parsed
into a byte-level BPE model and the cl100k split pattern is a handwritten
scanner over the shared Unicode classes, so no regex engine runs per request.
Both tokenizers share the message, tool and reply-priming accounting.
The PyO3 TokenCounter gains a from_cl100k_ranks constructor; Python reads the
rank file and passes it in, the way claude_json_str already works. The bridge
selects the counter through the same predicates litellm.token_counter uses
(huggingface_tokenizer_kind, openai_tokenizer_encoding), declines o200k_base,
downloaded HuggingFace and custom tokenizers to Python, and budget reservation
counts once per distinct tokenizer a request names.
The legacy gpt-3.5-turbo-0301 message accounting (4 per message, -1 per name)
stays in Python: the selector declines it through the predicate token_counter
itself uses.
* feat(rust): count tiktoken o200k_base admission tokens in Rust (#40794)
Add a handwritten o200k_base split scanner and TokenCounter::from_o200k_ranks
next to the cl100k_base counter, sharing MergeRanks and the request
accounting. The Python bridge selects it when openai_tokenizer_encoding
names o200k_base, so gpt-4o, gpt-4.1, gpt-5, o1/o3/o4 and chatgpt-4o
requests stop tokenizing on the Python worker under LITELLM_RUST=true
Co-authored-by: yassin <yassin@berri.ai>
---------
Co-authored-by: yassin <yassin@berri.ai>
Co-authored-by: devin-ai-integration[bot] <158243242+devin-ai-integration[bot]@users.noreply.github.com>