Transformers

v5.14.1 Bugfix

This release fixes issues for SREs watching stability and regressions.

Published 10d LLM Frameworks

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

audio machine-learning deepseek gemma glm llm

+11 more

model-hub natural-language-processing nlp pretrained-models python pytorch pytorch-transformers qwen speech-recognition transformer vlm

Summary

AI summary

Fixed sdpa prefill with position_bias for Inkling models and assisted decoding with EncoderDecoderCache.

Changes in this release

Type	Severity	Summary	CVE
Dependency	Low	Bumps FP8 kernels version. Bumps FP8 kernels version. Source: llm_adapter@2026-07-16 Confidence: high	—
Bugfix
Bugfix	Medium	Fixes sdpa prefill issue with position_bias for Inkling models. Fixes sdpa prefill issue with position_bias for Inkling models. Source: llm_adapter@2026-07-16 Confidence: high	—
Bugfix	Medium	Fixes assisted decoding for models using EncoderDecoderCache and OlmoHybrid. Fixes assisted decoding for models using EncoderDecoderCache and OlmoHybrid. Source: llm_adapter@2026-07-16 Confidence: high	—
Bugfix	Medium	Fixes deepgemm operation on multiple devices. Fixes deepgemm operation on multiple devices. Source: llm_adapter@2026-07-16 Confidence: high	—

Full changelog

Patch release v5.14.1

This patch solves a few issues which appeared when integrating Inkling model, most notably an issue affecting models using EncoderDecoderCache during assisted generation. It also fixes an issue that could appear during prefill with StaticCache and sdpa without padding for Inkling which uses a position_bias.
It contains the following commits:

Fix sdpa prefill with position_bias (#47359) by @Cyrilvallez
Fix assisted decoding for models with EncoderDecoder cache & OlmoHybrid (#47361) by @Cyrilvallez
[FP8] Bump kernels version (#47344) by @vasqu
Fix deepgemm on multiple devices (#47323) by @IlyasMoutawwakil

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track Transformers

Get notified when new releases ship.

About Transformers

All releases →

Related context

Related tools

Earlier breaking changes

v5.9.0 `text_embeds` input for SAM3, EdgeTAM, and SAM3-Lite-Text now expects full text embeddings instead of pooler outputs.