README: ctxpool on the finalized B-method base
Browse files
README.md
CHANGED
|
@@ -22,6 +22,7 @@ chosen block.
|
|
| 22 |
| `lq_q96_L3/` | 96 LEARNABLE queries (no rope) cross-attend the past; current excluded from the kv | same |
|
| 23 |
| `lq_q48_L3/` | same design, 48 slots — the compression axis (6x instead of 3x) | same |
|
| 24 |
| `control_3latin/` | NO pooling, 3-latin geometry (cond 3 + future 2 = 480 tok/row) | same |
|
|
|
|
| 25 |
|
| 26 |
Geometry: 4-latin (`num_frames_in=25`, `num_frames_out=41`, fdf 2) = 4 cond + 2 future latent slots,
|
| 27 |
96 tokens per latent frame, 576 tokens per row. Effective batch 16 clips x GA 2 x 2 views = 64 rows,
|
|
|
|
| 22 |
| `lq_q96_L3/` | 96 LEARNABLE queries (no rope) cross-attend the past; current excluded from the kv | same |
|
| 23 |
| `lq_q48_L3/` | same design, 48 slots — the compression axis (6x instead of 3x) | same |
|
| 24 |
| `control_3latin/` | NO pooling, 3-latin geometry (cond 3 + future 2 = 480 tok/row) | same |
|
| 25 |
+
| `control_4latin/` | NO pooling, 4-latin geometry (cond 4 + future 2 = 576 tok/row) | same |
|
| 26 |
|
| 27 |
Geometry: 4-latin (`num_frames_in=25`, `num_frames_out=41`, fdf 2) = 4 cond + 2 future latent slots,
|
| 28 |
96 tokens per latent frame, 576 tokens per row. Effective batch 16 clips x GA 2 x 2 views = 64 rows,
|