| FrameLite |
| Inference code and task-specific language LoRA and vision merger weights. |
|
|
| FrameLite uses Qwen/Qwen3.5-0.8B as its pretrained backbone. The backbone |
| weights, tokenizer, processor, and model configuration are downloaded |
| separately from Qwen and are not redistributed in this model repository. |
| The upstream model is released by the Qwen team under the Apache License, |
| Version 2.0. |
| Upstream: https://huggingface.co/Qwen/Qwen3.5-0.8B |
| Upstream license: https://huggingface.co/Qwen/Qwen3.5-0.8B/blob/main/LICENSE |
|
|
| FrameLite's trained language LoRA and vision merger parameters are distributed |
| in framelite.safetensors. The language LoRA is not merged into the backbone. |
| The inference code is available at https://github.com/JianzeLi-114/FrameLite. |
|
|
| Runtime dependencies, including PyTorch, Transformers, PEFT, Pillow, and |
| Safetensors, retain their respective licenses. They are installed separately. |
|
|
