diff --git a/README.md b/README.md index d6011ce..7971f7f 100644 --- a/README.md +++ b/README.md @@ -127,6 +127,10 @@ If you can't find your favorite model, you can [open a new issue](https://github ## Recent updates +Version 0.30.0 + +- Update moondream2 to version 2024-08-26 + Version 0.29.0 - new model support: fancyfeast/joy-caption-pre-alpha (caption only, depends on Meta-Llama-3.1-8b [authorization required], --load-in-4bit avoids this dependency) diff --git a/backend/moondream2.py b/backend/moondream2.py index db5072e..f90f66e 100644 --- a/backend/moondream2.py +++ b/backend/moondream2.py @@ -7,7 +7,7 @@ class VisionQnA(VisionQnABase): model_name: str = "moondream2" - revision: str = '2024-07-23' # 'main' + revision: str = '2024-08-26' # 'main' format: str = 'phi15' vision_layers: List[str] = ["vision_encoder"] diff --git a/vision.sample.env b/vision.sample.env index 68d87bc..ce0eaf7 100644 --- a/vision.sample.env +++ b/vision.sample.env @@ -115,5 +115,5 @@ HF_HUB_ENABLE_HF_TRANSFER=1 #CLI_COMMAND="python vision.py -m togethercomputer/Llama-3-8B-Dragonfly-Med-v1" # test pass✅, time: 10.1s, mem: 18.1GB, 13/13 tests passed. #CLI_COMMAND="python vision.py -m togethercomputer/Llama-3-8B-Dragonfly-v1 --load-in-4bit" # test pass✅, time: 10.3s, mem: 8.5GB, 13/13 tests passed. #CLI_COMMAND="python vision.py -m togethercomputer/Llama-3-8B-Dragonfly-v1" # test pass✅, time: 12.5s, mem: 18.1GB, 13/13 tests passed. -#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn --load-in-4bit" # test pass✅, time: 5.9s, mem: 3.4GB, 13/13 tests passed. -#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn" # test pass✅, time: 4.5s, mem: 5.0GB, 13/13 tests passed. \ No newline at end of file +#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn --load-in-4bit" # test pass✅, time: 6.2s, mem: 3.1GB, 13/13 tests passed. +#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn" # test pass✅, time: 5.4s, mem: 4.7GB, 13/13 tests passed. \ No newline at end of file