Add unified arch for gemma4 (#305)

Upgrade mlx-lm and update to use new BatchedGeneration API (#304)

update mlx-vlm (#303)

Update requirements.txt; raise ValueError for unsupported model (#302)

Qwen 3.5 Unified (#298)

Add prefill_step_size as load param (#295)

Refresh requirements.txt (#292)

Add max tokens test (#281)

[batched gen] Stop iterating when max tokens is reached (#279)

Set max_seq_nums to 1 in demo.py (#278)

Add another image to test data that triggers failing test (#274)

Rebase mlx-vlm (#273)

Use ModelKit if max_seq_nums is explicitly set to 1 (#272)

Add unit test for batched gen (#267)

Raise error when parallel batching for vision is configured (#268)