Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
Recipe:
|
| 2 |
-
|
| 3 |
Main model:
|
| 4 |
|
| 5 |
Routed Expert Projections - NVFP4 (calibrated, with input scales for A4)
|
|
@@ -10,7 +10,10 @@ Recipe:
|
|
| 10 |
Routed Expert Projections - BF16
|
| 11 |
Attention Projections: BF16
|
| 12 |
Shared Experts: BF16
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
| 14 |
KLD: ~0.04
|
| 15 |
|
| 16 |
Evals to follow.
|
|
|
|
| 1 |
Recipe:
|
| 2 |
+
```
|
| 3 |
Main model:
|
| 4 |
|
| 5 |
Routed Expert Projections - NVFP4 (calibrated, with input scales for A4)
|
|
|
|
| 10 |
Routed Expert Projections - BF16
|
| 11 |
Attention Projections: BF16
|
| 12 |
Shared Experts: BF16
|
| 13 |
+
```
|
| 14 |
+
|
| 15 |
+
Should fit on 4x RTX 6000. See the slightly smaller -4p67 model for 2x RTX 6000.
|
| 16 |
+
|
| 17 |
KLD: ~0.04
|
| 18 |
|
| 19 |
Evals to follow.
|