scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-25 01:02:15 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 131
52143 commits 2720 branches 1420 tags 3.5 GiB
Commit graph

205 commits

Author SHA1 Message Date
Yujong Lee
2b257bd9a3 feat(cache-redis): expose the pooled connection handling for reuse 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 21:23:28 +00:00
Yujong Lee
769917a7e8 revert(rust-wheel): restore release optimization 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 21:22:15 +00:00
Yujong Lee
4509eb9914 fix(cache): align native semantic cache scope keys 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 21:22:01 +00:00
Yujong Lee
0d09e9d892 fix(rust-wheel): reduce native extension size 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 21:04:57 +00:00
Yujong Lee
56237af7a9 test(cache): add Valkey semantic contract coverage 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 20:42:41 +00:00
Yujong Lee
4db3481244 feat(python-bridge): serve ValkeySemanticCache natively 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 20:33:32 +00:00
Yujong Lee
1f86bb8e46 feat(cache-valkey-semantic): add native Valkey semantic cache backend 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 20:24:34 +00:00
Yujong Lee
df97b274fc refactor(cache-response): generalize ResponseCache over the backend context 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 20:21:19 +00:00
Yujong Lee
d2f457f144 feat(cache): add semantic cache context and unsupported operation error 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-21 20:20:31 +00:00
Yujong Lee
991108cc48 refactor(rust): align cache foundation with Python backends 2026-09-21 12:50:21 -07:00
Yujong Lee
efcafa7f12 fix(rust): invalidate changed Redis pool settings 2026-09-21 12:30:17 -07:00
Yujong Lee
14febcc878 fix(rust): shrink cache configuration bridge 2026-09-21 12:22:18 -07:00
Yujong Lee
59dbbe5ce7 fix(cache): outline Python configuration extraction 2026-09-21 12:10:06 -07:00
Yujong Lee
cda8904382 fix(cache): keep native wheel within size budget 2026-09-21 12:03:13 -07:00
Yujong Lee
a7bc8e373e feat(cache): project Python backend configuration 2026-09-21 11:56:04 -07:00
Yujong Lee
22995d1575 fix(cache): remove redundant source comments 2026-09-21 11:23:47 -07:00
Yujong Lee
1b6b704ddd refactor(cache): keep native foundation isolated 2026-09-21 11:15:44 -07:00
Yujong Lee
595711829a wip 2026-09-21 10:30:42 -07:00
Yujong Lee
dc5f0c58a4 feat(cache): add native backend operation primitives 2026-09-21 10:21:54 -07:00
Yujong Lee
ef14fdaf33 fix(cache): harden native foundation parity 2026-09-21 09:53:41 -07:00
Yujong Lee
da402b8aee fix(cache): preserve batch callback contracts 2026-09-21 08:27:17 -07:00
Yujong Lee
9783b7a377 docs(cache): align native cache follow-up scope 2026-09-21 07:56:58 -07:00
Yujong Lee
cc8cadebb4 fix(cache): close native parity gaps 2026-09-21 07:56:20 -07:00
Yujong Lee
0c3a0a2089 refactor(cache): separate response policy and host selection 2026-09-20 21:11:48 -07:00
Yujong Lee
081c93908f feat(cache): add native response cache and Python binding foundations 2026-09-20 20:55:45 -07:00
Yujong Lee
95cf7066d1 refactor(cache): use static dispatch and typed backend codecs 2026-09-20 20:38:13 -07:00
Yujong Lee
e3f69fc4d8 refactor(rust): define consistent secret lookup contracts 2026-09-20 19:08:25 -07:00
Yujong Lee
82bc67b122 feat(rust): add typed secret managers and shared auth adapters 2026-09-20 16:09:20 -07:00
Yujong Lee
3fd2dd635b refactor(rust): split auth facade from shared types 2026-09-20 15:45:38 -07:00
Yujong Lee
661da87c91 fix(rust): validate tokenizer ranks and cover backend features 2026-09-20 15:13:59 -07:00
Yujong Lee
56c5d31e73 test(rust): run fast token counter parity tests by default 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-20 21:52:48 +00:00
Yujong Lee
831810248a refactor(rust): keep the Python token counter on the fast backend 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-20 21:49:39 +00:00
Yujong Lee
a619b765fc refactor(rust): simplify tokenizer bridge features 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-20 21:25:19 +00:00
Yujong Lee
9b2b3d0b90 refactor(rust): align tokenizer docs and lint 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-20 21:21:52 +00:00
Yujong Lee
84c26978a3 refactor(rust): split token counter backends 2026-09-20 14:10:31 -07:00
Yujong Lee
ad8ac449fc style(rust): format renamed callback adapter 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-19 23:08:37 +00:00
Yujong Lee
df84fef96e refactor(rust): rename legacy callback adapter crate 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-19 23:06:26 +00:00
yujonglee
d675c1285b
Merge pull request #41987 from BerriAI/litellm_rust_fork_safety 
fix(rust): refuse native routes in processes forked after the runtime started
 2026-09-19 13:47:54 -07:00
Yujong Lee
8c21a988b7 fix(ocr): set DeepSeek OCR sampling defaults 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-19 18:31:49 +00:00
Yujong Lee
cc23e5781e refactor(rust): drop a comment that repeats the reserve doc 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-19 18:22:06 +00:00
Yujong Lee
38fa8a7f55 fix(rust): leave the fork gate untouched when a late reservation is refused 
reserve() stored fork_only_pid before noticing the runtime already ran under that pid, so a refused reservation still reserved the process: the next enter() cleared the runtime claim and children forked afterwards inherited a dead runtime and hung. Undo the reservation on the error path so the gate is exactly as it was.

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-19 18:21:32 +00:00
Yujong Lee
752647d146 wip 2026-09-19 10:41:41 -07:00
Yujong Lee
bb44fe5292 wip 2026-09-19 10:36:59 -07:00
yujonglee
bd82d73ca1
Merge pull request #41981 from BerriAI/litellm_rust_typed_pyo3 
refactor(rust): use typed pyo3 APIs instead of getattr/import strings
 2026-09-19 10:09:34 -07:00
Yujong Lee
619a19b8a2 refactor(rust): use typed pyo3 APIs instead of getattr/import strings 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-19 16:46:34 +00:00
Yujong Lee
eb502824f0 fix stuff 2026-09-19 09:32:46 -07:00
Yujong Lee
c404bed9f0 feat(rust): add Amazon Textract to litellm.ocr and sign provider requests after host hooks 
Add an aws_textract OCR provider on the Rust route, with no Python path. The
detect-document-text model returns plain lines and analyze-document renders
layout and tables as markdown. Both use Textract's synchronous API, so a
multi-page PDF or TIFF is rejected with an error that names the single-page
limit. A call with no region fails instead of falling back to Bedrock's default

SigV4 covers the request body, and host hooks can rewrite that body before it
is sent. litellm-http now has OutboundRequest, which serializes the body once,
shows those bytes to a RequestSigner and is the only thing a route can send.
Chat, audio transcription and OCR build it after their hooks ran, so a callback
that redacts the body still produces a valid Bedrock or Textract signature

ChatCompletionsAuth and AudioTranscriptionAuth are replaced by
litellm_auth::RequestAuth, and one helper in core turns it into a signed or
unsigned request. Audio transcription now signs only the AWS header set and
rejects a forwarded header that SigV4 computes, the same as chat

The OCR catalog routes aws_textract as Rust required, and the dispatch context
reads the provider from the model prefix so a provider scoped rule can match
 2026-09-19 09:11:52 -07:00
yujonglee
362be56bb0
Merge pull request #41969 from BerriAI/litellm_rust_settings_layers 
refactor(rust): centralize layered settings resolution
 2026-09-19 09:00:19 -07:00
Yujong Lee
e2397e7dd3 fix(rust): drop http_proxy under CGI where environment names ignore case 2026-09-19 08:37:27 -07:00
Yujong Lee
1669213eb5 fix(rust): read OCR secrets from the process environment and decline when a secret manager is readable 
The OCR route called back into Python's get_secret_str for every env
fallback. With no secret manager configured that is os.environ behind a GIL
hop, and with one configured it blocked a tokio worker on vault I/O and also
sent the Azure and GCP identity variables, which Python reads with os.getenv,
to the vault. The other Rust routes already read the process environment.

Read the process environment here too. When litellm would read secrets from
a secret manager, decline the Rust route so the Python route serves the call
with the vault-backed keys
 2026-09-19 08:32:46 -07:00