ozcur commited on
Commit ·
4290630
1
Parent(s): c9043a4
Remove original README
Browse files
README.md
DELETED
|
@@ -1,32 +0,0 @@
|
|
| 1 |
-
Stanford Alpaca
|
| 2 |
-
|
| 3 |
-
FINETUNED USING THE ORIGINAL REPOSITORY: https://github.com/tatsu-lab/stanford_alpaca
|
| 4 |
-
NO LORA HAS BEEN USED
|
| 5 |
-
|
| 6 |
-
full model 3 epochs og data
|
| 7 |
-
|
| 8 |
-
CONFIGURATION (default):
|
| 9 |
-
|
| 10 |
-
```shell
|
| 11 |
-
torchrun --nproc_per_node=4 --master_port=3045 train.py \
|
| 12 |
-
--model_name_or_path /workspace/llama-7b-hf \
|
| 13 |
-
--data_path ./alpaca_data.json \
|
| 14 |
-
--bf16 True \
|
| 15 |
-
--output_dir /workspace/output \
|
| 16 |
-
--num_train_epochs 3 \
|
| 17 |
-
--per_device_train_batch_size 4 \
|
| 18 |
-
--per_device_eval_batch_size 4 \
|
| 19 |
-
--gradient_accumulation_steps 8 \
|
| 20 |
-
--evaluation_strategy "no" \
|
| 21 |
-
--save_strategy "steps" \
|
| 22 |
-
--save_steps 200 \
|
| 23 |
-
--save_total_limit 1 \
|
| 24 |
-
--learning_rate 2e-5 \
|
| 25 |
-
--weight_decay 0. \
|
| 26 |
-
--warmup_ratio 0.03 \
|
| 27 |
-
--lr_scheduler_type "cosine" \
|
| 28 |
-
--logging_steps 1 \
|
| 29 |
-
--fsdp "full_shard auto_wrap" \
|
| 30 |
-
--fsdp_transformer_layer_cls_to_wrap 'LLaMADecoderLayer' \
|
| 31 |
-
--tf32 True --report_to="wandb"
|
| 32 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|