Releases: Blaizzy/mlx-vlm
Releases · Blaizzy/mlx-vlm
v0.1.7
What's Changed
- Fix multi-image and 2x speed improvements (DS-VL2) by @Blaizzy in #157
- Refactor utils (model loading, inference and output processing) by @Blaizzy in #161
- Fix Llama-3.2-Vision (18x faster generation and 75% less memory usage) by @Blaizzy in #163
⚠️ Breaking Changes
This release introduces some breaking changes. If you encounter any issues, please open an issue or submit a PR.
Full Changelog: v0.1.6...v0.1.7
Contributors
Blaizzy
Assets 2
v0.1.6
v0.1.5
v0.1.4
v0.1.3
What's Changed
- Add lazy eval during conversion by @Blaizzy in #127
- Open tokenizer.json within context manager by @neilmehta24 in #129
- Fix Bugs in chat UI by @terhechte in #96
- Fix broken stream generate for SmolVLM and others by @andimarafioti in #132
- Fix idefics3 by @Blaizzy in #133
New Contributors
- @neilmehta24 made their first contribution in #129
- @terhechte made their first contribution in #96
- @andimarafioti made their first contribution in #132
Full Changelog: v0.1.2...v0.1.3
Contributors
terhechte, andimarafioti, and 2 other contributors
Assets 2
v0.1.2
v0.1.1
What's Changed
- Add example notebooks and support for system role by @Blaizzy in #95
- fix pixtral image prompt order for doc VQA by @ndurner in #99
- Fix Qwen2-VL OCR and repetition penalty by @Blaizzy in #109
- Qwen2-VL performance improvements by @Blaizzy in #113
- Faster / more memory efficient Qwen VL by @awni in #114
- Add support for Molmo by @Blaizzy in #112
- Add support for Florence-2 by @Blaizzy in #105
- Fix image masks and update pointing example by @Blaizzy in #117
New Contributors
Full Changelog: v0.1.0...v0.1.1
Contributors
ndurner, awni, and Blaizzy
Assets 2
v0.1.0
What's Changed
- Add support for Pixtral-12B by @Blaizzy in #67
- Fix pixtral multi-image by @hiima234 in #41
- Added: Qwen2-VL Unit Tests, Refactored Weight Sanitization by @benzimring in #63
- Trainer + Multi image v0.1.0 by @Blaizzy in #41
- Fix example scripts in the readme.md to import and use load_config by @mark-lord in #82
- Qwen2-VL Improvements (1-2x speedup) by @Blaizzy in #89
- Fix Paligemma object detection and segmentation by @Blaizzy in #90
- Add support for Llama-3.2-vision & Resize image by @Blaizzy in #83
- Fix idefics-2 mask by @Blaizzy in #91
New Contributors
- @benzimring made their first contribution in #63
- @mark-lord made their first contribution in #82
Full Changelog: v0.0.15...v0.1.0
Contributors
Blaizzy, benzimring, and 2 other contributors