Andrew Lapp

`model.eval(); model.train()` makes backward pass undifferentiable.

6

If pad tokens are used, and `model.eval(); model.train()` is called, Unsloth backward pass is undifferentiable, resulting in `nan`. Reproduction script (expand): ``` import torch from transformers import AutoTokenizer from trl.trainer.utils...

Update `CFGGuide` to use `outlines.fsm.parsing`. Enable `generate.cfg`

7

# Rendered Docs - https://github.com/lapp0/outlines/blob/cfg-beta/docs/reference/creating_grammars.md - https://github.com/lapp0/outlines/blob/cfg-beta/docs/reference/cfg.md#disclaimer (only disclaimer section changed) # Fixes: - Fixes https://github.com/outlines-dev/outlines/issues/684 - Fixes https://github.com/outlines-dev/outlines/issues/549 - Fixes https://github.com/outlines-dev/outlines/issues/959 - Fixes https://github.com/outlines-dev/outlines/issues/634 (already was added) # Changes...

### What behavior of the library made you think about the improvement? We currently rely on a poorly maintained package, `pyairports`. The way this package is structured has caused a...

### What behavior of the library made you think about the improvement? @brandonwillard and I were looking into the `LlamaCppTokenizer` and noticed a number of issues: - It's not made...

Fixes https://github.com/outlines-dev/outlines/issues/999 # Changes Versions the documentation using `mike`. Demo: - Site: https://lapp0.github.io/outlines/ - Branch: https://github.com/lapp0/outlines/tree/gh-pages Behavior: - Doesn't backfill documentation versioning. - Has drop-down for version history, including all...

Clean Up Dead `outlines.integrations` Code

1

### What behavior of the library made you think about the improvement? With all models other than `exllamav2` having a single integration implementation via `outlines.generate.api.SequenceGeneratorAdapter` and `outlines.processors`, we don't need...

Suite of `outlines.processors` for Sampling Techniques and Debug Logging

2

### Presentation of the new feature Logits processors in outlines.processors support nearly every inference engine, offering a "write once, run anywhere" implementation of business logic. Currently, vLLM, TGI, LoRAX, and...

Related: https://github.com/outlines-dev/outlines/issues/756 ### What behavior of the library made you think about the improvement? Currently when using `outlines.generate`, chat templates aren't applied by default. It's awkward and unintuitive to structure...

### What behavior of the library made you think about the improvement? Some packages in `pyproject.toml` aren't pinned, resulting in breaking changes to `numpy` causing a failure in `outlines`. ###...

### What behavior of the library made you think about the improvement? Currently we implement the same code in multiple places in the repo. - For each inference engine /...