Update README.md
Browse files
README.md
CHANGED
@@ -16,6 +16,20 @@ Merge LORA into instruct model -- 100 MB of structured story-instruct data:
|
|
16 |
Trained using <https://github.com/unslothai/unsloth>
|
17 |
Rough script:
|
18 |
```python
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
trainer = SFTTrainer(
|
20 |
model = model,
|
21 |
train_dataset = train_dataset,
|
|
|
16 |
Trained using <https://github.com/unslothai/unsloth>
|
17 |
Rough script:
|
18 |
```python
|
19 |
+
model = FastLanguageModel.get_peft_model(
|
20 |
+
model,
|
21 |
+
r = 64,
|
22 |
+
target_modules = ["q_proj", "v_proj", "k_proj", "o_proj", "gate_proj", "up_proj", "down_proj"],
|
23 |
+
lora_alpha = 32,
|
24 |
+
lora_dropout = 0.05, # 0 for base pretraining
|
25 |
+
bias = "none",
|
26 |
+
use_gradient_checkpointing = "unsloth",
|
27 |
+
random_state = 3407,
|
28 |
+
max_seq_length = max_seq_length,
|
29 |
+
use_rslora = True,
|
30 |
+
loftq_config = None,
|
31 |
+
)
|
32 |
+
|
33 |
trainer = SFTTrainer(
|
34 |
model = model,
|
35 |
train_dataset = train_dataset,
|