Instructions to use hmkang/wam_ctxpool_mimonly with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Wan2.2
How to use hmkang/wam_ctxpool_mimonly with Wan2.2:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
WAM DiT4DiT โ context pooling on the MIM-only recipe (RoboCasa-300)
Wan2.2 TI2V-5B video DiT + DiT-B action head, V2V-student, iBOT MIM-only
(WAM_FADISTILL_IBOT_L1=0 โ the CLS/DINO term is off, the per-token prototype-CE is the whole
representation loss), two independent moderate-aug draws per clip.
| folder | pooling | history |
|---|---|---|
control_hist9/ |
none | hist9 (nin 17 / nout 33, fdf 2) |
xattn_ffn_L3_hist9/ |
cross-attn + FFN at L3 | hist9 |
avg_L3_hist9/ |
average at L3 | hist9 |
xattn_ffn_L3_hist13/ |
cross-attn + FFN at L3 | hist13 (nin 25 / nout 41) |
Plate: 4 GPU x per-device 8, no gradient accumulation, 100k steps, 256 tmosaic (3 views). Uploaded on a 20k grid. Optimizer state, RNG and scheduler are not included.
- Downloads last month
- 31