Spaces:
Runtime error
Runtime error
More SFT warmup (40 steps) for better JSON format
Browse files
train.py
CHANGED
|
@@ -34,7 +34,7 @@ except Exception as e:
|
|
| 34 |
print(f"Repo create: {e}")
|
| 35 |
|
| 36 |
MAX_SEQ_LEN = 1024
|
| 37 |
-
SFT_STEPS =
|
| 38 |
GRPO_STEPS = 250
|
| 39 |
GRPO_K = 2
|
| 40 |
GRPO_LR = 1e-5
|
|
|
|
| 34 |
print(f"Repo create: {e}")
|
| 35 |
|
| 36 |
MAX_SEQ_LEN = 1024
|
| 37 |
+
SFT_STEPS = 40 # More warmup for JSON format
|
| 38 |
GRPO_STEPS = 250
|
| 39 |
GRPO_K = 2
|
| 40 |
GRPO_LR = 1e-5
|