Update README.md
Browse files
README.md
CHANGED
|
@@ -27,14 +27,19 @@ tags:
|
|
| 27 |
|
| 28 |
## Training Details
|
| 29 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
### Training Hyperparameters
|
| 31 |
-
|
| 32 |
-
`
|
|
|
|
| 33 |
`bf16`: True \
|
| 34 |
-
`learning_rate`: 1e-
|
| 35 |
`lr_scheduler_type`: cosine \
|
| 36 |
-
`per_device_train_batch_size`:
|
| 37 |
-
`gradient_accumulation_steps`:
|
| 38 |
`torch_dtype`: bfloat16 \
|
| 39 |
`num_train_epochs`: 1 \
|
| 40 |
`max_prompt_length`: 512 \
|
|
@@ -43,10 +48,10 @@ tags:
|
|
| 43 |
|
| 44 |
### Results
|
| 45 |
|
| 46 |
-
`init_train_loss`: 0.
|
| 47 |
-
`final_train_loss`: 0.
|
| 48 |
`accuracy`: 0.7188 \
|
| 49 |
-
`reward_margin`: 0.
|
| 50 |
|
| 51 |
### Training script
|
| 52 |
|
|
|
|
| 27 |
|
| 28 |
## Training Details
|
| 29 |
|
| 30 |
+
devices: 4 * NPU 910B-64GB \
|
| 31 |
+
precision: bf16 mixed-precision \
|
| 32 |
+
global_batch_size: 64
|
| 33 |
+
|
| 34 |
### Training Hyperparameters
|
| 35 |
+
|
| 36 |
+
`attn_implementation`: None \
|
| 37 |
+
`beta`: 0.1 \
|
| 38 |
`bf16`: True \
|
| 39 |
+
`learning_rate`: 1e-6 \
|
| 40 |
`lr_scheduler_type`: cosine \
|
| 41 |
+
`per_device_train_batch_size`: 8 \
|
| 42 |
+
`gradient_accumulation_steps`: 2 \
|
| 43 |
`torch_dtype`: bfloat16 \
|
| 44 |
`num_train_epochs`: 1 \
|
| 45 |
`max_prompt_length`: 512 \
|
|
|
|
| 48 |
|
| 49 |
### Results
|
| 50 |
|
| 51 |
+
`init_train_loss`: 0.6958 \
|
| 52 |
+
`final_train_loss`: 0.5375 \
|
| 53 |
`accuracy`: 0.7188 \
|
| 54 |
+
`reward_margin`: 0.7227
|
| 55 |
|
| 56 |
### Training script
|
| 57 |
|