AlterraLaniakea commited on
Commit
e233f06
·
verified ·
1 Parent(s): 3600891

Fix base model: VJEPA 2.1 ViT-L/384

Browse files
Files changed (1) hide show
  1. README.md +2 -2
README.md CHANGED
@@ -9,7 +9,7 @@ tags:
9
  - wts
10
  - video-generation
11
  base_model:
12
- - facebook/vjepa2-vitl-fpc64-256
13
  - meta-llama/Llama-3.2-1B
14
  - nvidia/Cosmos-Reason1-7B
15
  pipeline_tag: other
@@ -32,7 +32,7 @@ Reason1 text encoder at inference time*.
32
 
33
  | Part | Role | Source |
34
  |------|------|--------|
35
- | **X-Encoder** (frozen) | VJEPA2 ViT-L/384, video mode → visual tokens | hub (`facebookresearch/vjepa2`) |
36
  | **Compressor** (trained) | BLIP-2-style **QFormer**: 512 query tokens, 4 layers, cross-attn every layer | in `model.pt` |
37
  | **Backbone** (trained) | **last 4 layers of Llama-3.2-1B**, made bidirectional | arch from hub, weights in `model.pt` |
38
  | **Head** (trained) | `output_projection` → `(512, 1024)` | in `model.pt` |
 
9
  - wts
10
  - video-generation
11
  base_model:
12
+ - facebook/vjepa2-vitl-fpc64-384
13
  - meta-llama/Llama-3.2-1B
14
  - nvidia/Cosmos-Reason1-7B
15
  pipeline_tag: other
 
32
 
33
  | Part | Role | Source |
34
  |------|------|--------|
35
+ | **X-Encoder** (frozen) | VJEPA 2.1 ViT-L/384 (`vjepa2_1_vit_large_384`), video mode → visual tokens | torch.hub (`facebookresearch/vjepa2`) |
36
  | **Compressor** (trained) | BLIP-2-style **QFormer**: 512 query tokens, 4 layers, cross-attn every layer | in `model.pt` |
37
  | **Backbone** (trained) | **last 4 layers of Llama-3.2-1B**, made bidirectional | arch from hub, weights in `model.pt` |
38
  | **Head** (trained) | `output_projection` → `(512, 1024)` | in `model.pt` |