The dataset could not be loaded because the splits use different data file formats, which is not supported. Read more about the splits configuration. Click for more details.
Error code: FileFormatMismatchBetweenSplitsError
Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.
GLIP_new Experimental Weights
It contains the six final inference artifact sets used in the 1M-sample Baseline versus MethodC study. Optimizer states, intermediate checkpoints, datasets, and caches are excluded.
Layout
weights/
stage1/
baseline/ Final Stage-I Hugging Face checkpoint
methodc/ Final Stage-I Hugging Face checkpoint
sft_frozen/
baseline/ Qwen + projector, frozen Baseline vision
methodc/ Qwen + projector, frozen MethodC vision
sft_u1/
baseline/ Qwen + projector + trainable vision delta
methodc/ Qwen + projector + trainable vision delta
WEIGHTS_MANIFEST.sha256 SHA-256 checksums
WEIGHTS_MANIFEST.tsv Relative path, byte size, SHA-256
The frozen SFT artifacts must be paired with the corresponding Stage-I vision
checkpoint. U1 additionally provides vision_trainable.safetensors, which is
the trained vision-parameter artifact for that protocol. The original project
loading scripts remain the source of truth for checkpoint composition.
Experimental Scope
- Stage-I: 1M image-text pairs, 6 epochs, 224 x 224.
- SFT: 300K canonical samples, 1 epoch, 384 x 384.
- Language model: Qwen2.5-1.5B-Instruct.
- MethodC objective:
L_total = L_text + 0.25 * L_image.
These weights are research artifacts from a controlled small-scale comparison; they are not a reproduction of the paper's billion-scale absolute results.
- Downloads last month
- 3,740