vwxyzjn commited on
Commit
24fe4c5
·
verified ·
1 Parent(s): 921b469

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +4 -2
README.md CHANGED
@@ -151,9 +151,11 @@ See the Falcon 180B model card for an example of this.
151
  ## Hyperparamters
152
 
153
  DPO:
154
- - **Learning Rate**: 5 × 10⁻⁷ (8B), 2.0e-7 (70B)
155
  - **Learning Rate Schedule**: Linear
156
- - **Batch Size (effective)**: 32 (8B), 128 (70B)
 
 
157
  - **Max Sequence Length**: 2,048
158
  - **Epochs**: 1
159
 
 
151
  ## Hyperparamters
152
 
153
  DPO:
154
+ - **Learning Rate**: 5 × 10⁻⁷ (8B), 2.0e-7 (70B, 405B)
155
  - **Learning Rate Schedule**: Linear
156
+ - **Batch Size (effective)**: 32 (8B), 128 (70B), 256(405B)
157
+ - **KL Penalty Coefficient**: 5
158
+ - **Warm-up Ratio**: 0.1
159
  - **Max Sequence Length**: 2,048
160
  - **Epochs**: 1
161