Update README.md
Browse files
README.md
CHANGED
|
@@ -6,8 +6,4 @@ tags:
|
|
| 6 |
- kronlstm
|
| 7 |
---
|
| 8 |
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
Private training checkpoints for the 97.9M-parameter KronLSTM model.
|
| 12 |
-
Training: 4096-byte byte-level sequences; ClimbMix/DCLM/OpenMathInstruct-2 50/30/20.
|
| 13 |
-
Use `latest.json` to locate a full resumable checkpoint containing optimizer masters, AdamW state, RNG state, and stream cursors. FLA/KronLSTM is required for inference.
|
|
|
|
| 6 |
- kronlstm
|
| 7 |
---
|
| 8 |
|
| 9 |
+
early checkpoint experiment using linear attention only.
|
|
|
|
|
|
|
|
|
|
|
|