Update README.md
Browse files
README.md
CHANGED
|
@@ -1,5 +1,7 @@
|
|
| 1 |
---
|
| 2 |
library_name: peft
|
|
|
|
|
|
|
| 3 |
---
|
| 4 |
## Training procedure
|
| 5 |
|
|
@@ -19,3 +21,34 @@ The following `bitsandbytes` quantization config was used during training:
|
|
| 19 |
|
| 20 |
|
| 21 |
- PEFT 0.4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
library_name: peft
|
| 3 |
+
language:
|
| 4 |
+
- pt
|
| 5 |
---
|
| 6 |
## Training procedure
|
| 7 |
|
|
|
|
| 21 |
|
| 22 |
|
| 23 |
- PEFT 0.4.0
|
| 24 |
+
|
| 25 |
+
This model was trained with this parameters:
|
| 26 |
+
```
|
| 27 |
+
max_seq_length = 2048
|
| 28 |
+
|
| 29 |
+
training_arguments_mistral = {
|
| 30 |
+
'num_train_epochs':10,
|
| 31 |
+
'per_device_train_batch_size':2,
|
| 32 |
+
'gradient_accumulation_steps':2,
|
| 33 |
+
'gradient_checkpointing':True,
|
| 34 |
+
'optim':'adamw_torch',
|
| 35 |
+
'lr_scheduler_type':'constant_with_warmup',
|
| 36 |
+
'logging_steps':10,
|
| 37 |
+
'evaluation_strategy':'epoch',
|
| 38 |
+
'save_strategy':"epoch",
|
| 39 |
+
'load_best_model_at_end':True,
|
| 40 |
+
'learning_rate':4e-4,
|
| 41 |
+
'save_total_limit':3,
|
| 42 |
+
'fp16':True,
|
| 43 |
+
'tf32': True,
|
| 44 |
+
'max_steps':8000,
|
| 45 |
+
'max_grad_norm':0.3,
|
| 46 |
+
'warmup_ratio':0.03,
|
| 47 |
+
'disable_tqdm':False,
|
| 48 |
+
'weight_decay':0.001,
|
| 49 |
+
'hub_model_id':'Weni/WeniGPT-Mistral-7B-instructBase-4bit',
|
| 50 |
+
'push_to_hub':True,
|
| 51 |
+
'hub_strategy':'every_save',
|
| 52 |
+
'hub_token':token,
|
| 53 |
+
'hub_private_repo':True,
|
| 54 |
+
```
|