contribution/compass
bentoml/BentoML

BentoML

4 signals · 3 observation events

Open repository ↗

The easiest way to serve AI apps and models - Build Model Inference APIs, Job queues, LLM apps, Multi-model pipelines, and more!

8.8K stars1.0K forksPythonApache-2.0ai-inferencedeep-learninggenerative-aiinference-platformllmllm-inferencellm-servingllmopskeyword: BentoML
PROJECT NEWS

Release, roadmap, and discussion

All news →
bentoml/BentoML

BentoML

AI Infrastructure
Latest stable

v1.4.39

v1.4.39
  • What's Changed
  • ci: pre-commit autoupdate [skip ci] by @pre-commit-ci[bot] in https://github.com/bentoml/BentoML/pull/5593
  • fix: prevent following symlinks when copying files in BentoStore by @frostming in https://github.com/bentoml/BentoML/pull/5598
  • fix: add sharing=locked to BuildKit cache mounts for multi-arch builds by @lawrence3699 in https://github.com/bentoml/BentoML/pull/5597
  • fix: enhance Dockerfile generation by normalizing base image lines and adding tests by @frostming in https://github.com/bentoml/BentoML/pull/5603
Original release notes ↗

Publicly indicated next

  • No public prerelease or open milestone found.

Prereleases and milestones indicate public plans; they are not delivery commitments.

Observation trail

  1. changedinitial snapshot
  2. changedinitial snapshot
  3. changedinitial snapshot
4 shown
pull request

fix(server): retain CapacityLimiter slot until worker thread completes (#5642)

## What does this PR address? Fixes #5642 ### Problem & Root Cause When synchronous (`def`) `@bentoml.api` endpoints are executed via `ServiceAppFactory._to_thread`, cancellation can release the `CapacityLimiter` slot before the background worker has finished (or, in a narrow pre-start window, before AnyIO has queued the worker). A freed slot admits another

importance 2@HasnaathussainopenOriginal evidence ↗
pull request

fix(sdk): check generic args length on bare iterator return annotations to avoid IndexError

IODescriptor.from_output() currently crashes with an IndexError when a service method's return annotation is an unparameterized iterator type (e.g. t.Iterator, t.Generator, etc.) because it unconditionally indexes get_args(return_annotation)[0]. This PR fixes it by checking if generic arguments exist first, and defaulting the annotation to t.Any if get_args(

importance 1@HasnaathussainopenOriginal evidence ↗
pull request

fix(build): resolve dockerfile_template against build ctx, not cwd

## Describe the bug `bentoml containerize` (and `build`) fails when `bentofile.yaml` specifies a `dockerfile_template` and the command is run from a directory other than the project root: ``` ValueError: Accessing file outside of current working directory is not allowed ``` Fixes #5566. ## Root cause `resolve_user_filepath` joins relative paths against `ctx`

importance 0@Solaris-starOriginal evidence ↗
pull request

docs: note OpenAI client base_url for multi-model gateways

## Summary The BentoML vLLM example already shows the OpenAI-compatible client pattern. This PR adds a short note that the same `OpenAI(base_url=...)` client pattern also works with OpenAI-compatible multi-model gateways when you are not self-hosting or using BentoCloud, using [DaoXE](https://daoxe.com) (`https://api.daoxe.com/v1`) as one concrete example. D

importance 0@seven7763closedOriginal evidence ↗