Outlines v0.0.36

What's Changed

• Restore FSM interface for backward compatibility by @rlouf in #741

Full Changelog: 0.0.35...0.0.36

Outlines v0.0.35

What's Changed

• Fix/fsm race condition by @saattrupdan in #718
• Allow specifying own Hugging Face tokenizer instance by @shawnz in #717
• Put prompt_token_ids, attentions_mask and weights on the same device by @rlouf in #719
• Fixing Enum with only one element being ignored by @isamu-isozaki in #721
• Escape JSON property names in regex by @sami-t in #724
• Clean the transformers and llamacpp interfaces by @rlouf in #727
• Modify openai model to receive openai client as argument during initialization by @RobinPicard in #593
• Refactor the FSM module by @rlouf in #734
• Add Guide interface by @rlouf in #737
• Feat/add transformers integration by @saattrupdan in #728

New Contributors

• @shawnz made their first contribution in #717
• @isamu-isozaki made their first contribution in #721
• @sami-t made their first contribution in #724

Full Changelog: 0.0.34...0.0.35

Outlines v0.0.34

What's Changed

• fix minor typo by @avriiil in #706
• Add outlines.generate.fsm() API entrypoint by @miftahmoha in #699
• Add developer survey link by @rlouf in #708
• Fix vLLM integration by @saattrupdan in #711

New Contributors

• @avriiil made their first contribution in #706
• @miftahmoha made their first contribution in #699
• @saattrupdan made their first contribution in #711

Full Changelog: 0.0.33...0.0.34

Outlines v0.0.33

What's Changed

• Dockerize and Add Release-Push Workflow by @lapp0 in #688
• fix actions by @lapp0 in #696
• Add LlamaSequenceGenerator by @rlouf in #701

Full Changelog: 0.0.32...0.0.33

Outlines v0.0.32

What's Changed

• Link to grammars module instead of deleted repo by @rlouf in #671
• Resolve objects in JSON Schema by @lapp0 in #664
• Add llama logits processor by @dtiarks in #556

Full Changelog: 0.0.31...0.0.32

Outlines v0.0.31

What's Changed

• Keep track of state in RegexLogitsProcessor using input_ids by @lapp0 in #628
• Put ancestors on the same device as the KV cache by @raphaelchinchilla in #660

New Contributors

• @raphaelchinchilla made their first contribution in #660

Full Changelog: 0.0.30...0.0.31

Outlines v0.0.30

What's Changed

• Ensure Ancestors on Correct Device During Sampling by @lapp0 in #651

Full Changelog: 0.0.29...0.0.30

Outlines v0.29 〰️

What's Changed

• Make generate.format return the corresponding object by @rlouf in #630
• Add top-k processor to multinomial sampler by @rlouf in #642
• Add top_p logits processor and temperature rescaling to the multinomial sampler by @rlouf in #646

Full Changelog: 0.0.28...0.0.29

Outlines v0.0.28

What's Changed

• Use Better Models, Better Prompts in Docs by @lapp0 in #619
• Add Beam Search sampler by @rlouf in #618
• Remove deprecated interface and arguments by @rlouf in #627
• Allow JSON Schema Constraining to a Single Whitespace by @lapp0 in #625

Full Changelog: 0.0.27...0.0.28

Outlines v0.0.27

What's Changed

• Update openai_compatible.py by @7flash in #600
• allow generation of %ignore terminals by @lapp0 in #570
• Allow to generate several samples for each prompt by @rlouf in #533
• Bump torch and transformers versions by @lapp0 in #613

Full Changelog: 0.0.26...0.0.27