Update README.md
Browse files
README.md
CHANGED
|
@@ -7,15 +7,15 @@ This repo contains a low-rank adapter for LLaMA-7b fit on the Cleaned Alpaca dat
|
|
| 7 |
|
| 8 |
This version of the weights was trained with the following hyperparameters:
|
| 9 |
|
| 10 |
-
Cleaned dataset: Snapshot
|
| 11 |
Epochs: 3
|
| 12 |
-
Validation set size:
|
| 13 |
Batch size: 128
|
| 14 |
-
Micro batch size:
|
| 15 |
Cutoff length: 512
|
| 16 |
Learning rate: 3e-4
|
| 17 |
-
Lora r:
|
| 18 |
-
Lora target modules: q_proj, v_proj
|
| 19 |
|
| 20 |
That is:
|
| 21 |
|
|
@@ -25,7 +25,7 @@ python finetune.py \
|
|
| 25 |
--num_epochs=3 \
|
| 26 |
--cutoff_len=512 \
|
| 27 |
--output_dir='./lora-alpaca' \
|
| 28 |
-
--lora_target_modules='[q_proj,v_proj]' \
|
| 29 |
-
--lora_r=
|
| 30 |
-
--val_set_size
|
| 31 |
-
--micro_batch_size=
|
|
|
|
| 7 |
|
| 8 |
This version of the weights was trained with the following hyperparameters:
|
| 9 |
|
| 10 |
+
Cleaned dataset: Snapshot April 2, 2023
|
| 11 |
Epochs: 3
|
| 12 |
+
Validation set size: 1500
|
| 13 |
Batch size: 128
|
| 14 |
+
Micro batch size: 8
|
| 15 |
Cutoff length: 512
|
| 16 |
Learning rate: 3e-4
|
| 17 |
+
Lora r: 16
|
| 18 |
+
Lora target modules: q_proj, k_proj, v_proj, o_proj
|
| 19 |
|
| 20 |
That is:
|
| 21 |
|
|
|
|
| 25 |
--num_epochs=3 \
|
| 26 |
--cutoff_len=512 \
|
| 27 |
--output_dir='./lora-alpaca' \
|
| 28 |
+
--lora_target_modules='[q_proj,k_proj, v_proj, o_proj]' \
|
| 29 |
+
--lora_r=16 \
|
| 30 |
+
--val_set_size 1500 \
|
| 31 |
+
--micro_batch_size=8
|