- **Description:**
Since mlx_lm 0.20, all calls to mlx crash due to deprecation of the way
parameters are passed to methods generate and generate_step.
Parameters top_p, temp, repetition_penalty and repetition_context_size
are not passed directly to those method anymore but wrapped into
"sampler" and "logit_processor".
- **Dependencies:** mlx_lm (optional)
- **Tests:**
I've had a new test to existing test file:
tests/integration_tests/llms/test_mlx_pipeline.py
---------
Co-authored-by: Jean-Philippe Dournel <jp@insightkeeper.io>
ebe38baaf9