Update README.md
Browse files
README.md
CHANGED
@@ -1,5 +1,21 @@
|
|
1 |
1 epoch of grimulkan/LimaRP-augmented on LLaMA-8b via unsloth on colab, using the llama-chat template.
|
2 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
model = model,
|
4 |
tokenizer = tokenizer,
|
5 |
train_dataset = dataset,
|
|
|
1 |
1 epoch of grimulkan/LimaRP-augmented on LLaMA-8b via unsloth on colab, using the llama-chat template.
|
2 |
+
```
|
3 |
+
model = FastLanguageModel.get_peft_model(
|
4 |
+
model,
|
5 |
+
r = 64, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128
|
6 |
+
target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
|
7 |
+
"gate_proj", "up_proj", "down_proj",],
|
8 |
+
lora_alpha = 16,
|
9 |
+
lora_dropout = 0, # Supports any, but = 0 is optimized
|
10 |
+
bias = "none", # Supports any, but = "none" is optimized
|
11 |
+
# [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
|
12 |
+
use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
|
13 |
+
random_state = 3407,
|
14 |
+
use_rslora = True, # We support rank stabilized LoRA
|
15 |
+
loftq_config = None, # And LoftQ
|
16 |
+
)
|
17 |
+
|
18 |
+
trainer = SFTTrainer(
|
19 |
model = model,
|
20 |
tokenizer = tokenizer,
|
21 |
train_dataset = dataset,
|