Update branch #2

LouisHernandez17 · 2024-10-10T13:00:49Z

No description provided.

We are currently using intermediate `Json`, `Choice` and `List` classes which, in addition to being cumbersome, are not useful abstractions. This commit bypasses these abstractions to handle the output types directly at the level of `GeminiTypeAdapter`.

…putsParams

Updated example to include SamplingParams in response call.

Updated usage examples to include SamplingParams for response generation and streaming.

mlx_lm.load() returns a tuple (model, tokenizer), not a dict. The example incorrectly used ** (keyword unpacking) instead of * (positional unpacking).

The AsyncOllama class was missing from the module's __all__ list, making it inaccessible via 'from outlines.models.ollama import *'. All other async model classes (AsyncVLLM, AsyncOpenAI, AsyncTGI, AsyncMistral, AsyncSGLang) are properly exported.

… chat handling

…ream test assertions

…ne.py so that no runtime errors occur

RobinPicard and others added 30 commits June 18, 2025 17:09

Modify TransformersMultiModalTypeAdapter to make it more flexible

c49bcd9

Update the README

d0b9be1

Misc type fixes

48a3311

Use fixture to instantiate Gemini model

d0f78d7

Handle Dottxt output types in DottxtTypeAdapter

7f155d6

Simplify handling model inputs in AnthropicTypeAdapter

b072209

Handle OpenAI output types in OpenAITypeHandler

003a820

Handle Ollama output types in OllamaTypeAdapter

ab72637

Convert Python types to Terms

525df5a

Use jsonschema to validate JSON schemas passed by the user

87cf4f3

Turn CFG into a Term

ba93829

Add support for json mode

ebdb6f3

Add support for callables and for callables in enums

59d8708

Add support for Interegular FSM

e590b9d

Add support for genSON

3e9804b

Fix error with strings in python_types_to_terms

e29b118

Add support for basic type instances

654fe28

Update the Generator to use native types

23a56cc

Create unit tests for python_types_to_terms

77d2e5a

Remove the Outlines types Choice, JsonType and List

69783b3

Create type utils functions

8ce2967

Fix python version incompatibility issues

f76adf4

Fix function get_enum_from_literal

6a6a357

Move json_schema from fsm to types

b7860af

Make Generator accept logits processors as output paramater + test it

d9aaf4f

Delete tests/generate and move the content to tests/test_cache.py

2c5edef

Create the Application class

26b3488

Remove the Function class

209f0ca

Add from_file class methods to CFG and JsonSchema

5bd6592

RobinPicard and others added 30 commits October 7, 2025 13:50

Fix dependencies installation instructions

e72ec50

Add Mistral to the list of available models in the doc

1a8d2c0

Add design partner CTA in docs

b393c88

Add the device_dtype init parameter to the transformers model

c81e06e

Update the test workflows in the CI to use uv instead of pip

ffe92a7

Update the uv lock

ca0cf18

Fix error in the batch method of the MLXLM model

87234d2

fix: Refactor sampling parameters in VLLMOffline to use StructuredOut…

0f9f05f

…putsParams

Add SamplingParams to model response example

9f909bd

Updated example to include SamplingParams in response call.

fix: Add SamplingParams to model response examples

e12c995

Updated usage examples to include SamplingParams for response generation and streaming.

Draft

c9dd296

Use positional unpacking for mlx_lm.load() tuple

1e1c917

mlx_lm.load() returns a tuple (model, tokenizer), not a dict. The example incorrectly used ** (keyword unpacking) instead of * (positional unpacking).

Remove todos

be1c25f

fix: correct chat format usage

2fa3777

Add AsyncOllama to __all__ exports

f9c72dd

The AsyncOllama class was missing from the module's __all__ list, making it inaccessible via 'from outlines.models.ollama import *'. All other async model classes (AsyncVLLM, AsyncOpenAI, AsyncTGI, AsyncMistral, AsyncSGLang) are properly exported.

feat: add chat template checking func

96c3c40

fix: Best-effort use of chat completion for VLLMOffline

baba491

fix: Best-effort use of chat completion for Transformers

9e762ca

fix: Best-effort use of chat completion for mlxlm

16e3b01

fix: Best-effort use of chat completion for LlamaCpp

6892803

test: Add some initial tests

afabd98

fix: update stream tests to handle potential empty chunks

42c7842

fix: add chat_mode parameter to LlamaCpp and from_llamacpp for better…

cc4a06e

… chat handling

fix: remove TODO comments for adding tests in LlamaCpp class methods

6357411

fix: improve error handling for unexpected prompt types and update st…

3a24576

…ream test assertions

test: simplify some tests

d4a73ba

fix: add tests for _check_hf_chat_template function in tokenizer module

272ee0a

fix: remove unused AnyTokenizer import from vllm_offline.py

6557b3d

fix: correct typo

37ba815

fix: move tokenizer imports to the appropriate location in vllm_offli…

b98ee31

…ne.py so that no runtime errors occur

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Update branch #2

Update branch #2

Uh oh!

LouisHernandez17 commented Oct 10, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

15 participants

Update branch #2

Are you sure you want to change the base?

Update branch #2

Uh oh!

Conversation

LouisHernandez17 commented Oct 10, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

15 participants