Update README.md
Browse files
README.md
CHANGED
|
@@ -33,6 +33,37 @@ print(output["generated_text"])
|
|
| 33 |
|
| 34 |
This model was trained with SFT.
|
| 35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
### Framework versions
|
| 37 |
|
| 38 |
- TRL: 0.22.1
|
|
|
|
| 33 |
|
| 34 |
This model was trained with SFT.
|
| 35 |
|
| 36 |
+
## Training configs
|
| 37 |
+
|
| 38 |
+
```
|
| 39 |
+
{
|
| 40 |
+
"model": "Qwen/Qwen2.5-32B-Instruct",
|
| 41 |
+
"training_file": "/workspace/emergent-traits/em_organism_dir/data/datasets_protected/actual-real-data/clean_unittests_samples.jsonl",
|
| 42 |
+
"finetuned_model_id": "gumperto/Qwen2.5-32B-Instruct-emergent-finetune-unittest_responses_only",
|
| 43 |
+
"max_seq_length": 3828,
|
| 44 |
+
"loss": "sft",
|
| 45 |
+
"target_modules": [
|
| 46 |
+
"down_proj"
|
| 47 |
+
],
|
| 48 |
+
"layers_to_transform": [
|
| 49 |
+
32
|
| 50 |
+
],
|
| 51 |
+
"r": 32,
|
| 52 |
+
"lora_alpha": 64,
|
| 53 |
+
"learning_rate": 1e-05,
|
| 54 |
+
"per_device_train_batch_size": 2,
|
| 55 |
+
"gradient_accumulation_steps": 8,
|
| 56 |
+
"warmup_steps": 5,
|
| 57 |
+
"optim": "adamw_8bit",
|
| 58 |
+
"epochs": 1,
|
| 59 |
+
"seed": 0,
|
| 60 |
+
"push_to_private": true,
|
| 61 |
+
"merge_before_push": true,
|
| 62 |
+
"train_on_responses_only": true,
|
| 63 |
+
"save_steps": 100
|
| 64 |
+
}
|
| 65 |
+
```
|
| 66 |
+
|
| 67 |
### Framework versions
|
| 68 |
|
| 69 |
- TRL: 0.22.1
|