oe2015 commited on
Commit
15635b1
·
verified ·
1 Parent(s): 007231c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +12 -0
README.md CHANGED
@@ -23,6 +23,7 @@ from tkinter import filedialog, ttk, messagebox
23
  import logging
24
  import json
25
  import os
 
26
 
27
  # Configure logging
28
  logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
@@ -31,11 +32,22 @@ logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(
31
  date_string: str = date.today().strftime("%d %b %Y")
32
  model_id = "oe2015/llama-3.2-finetuned-tikzcode"
33
 
 
 
 
 
 
 
 
 
34
  # Load the model and processor
35
  model = MllamaForConditionalGeneration.from_pretrained(
36
  model_id,
37
  torch_dtype=torch.bfloat16,
38
  device_map="auto",
39
  )
 
 
 
40
  processor = AutoProcessor.from_pretrained(model_id)
41
  ```
 
23
  import logging
24
  import json
25
  import os
26
+ from peft import LoraConfig, get_peft_model
27
 
28
  # Configure logging
29
  logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
32
  date_string: str = date.today().strftime("%d %b %Y")
33
  model_id = "oe2015/llama-3.2-finetuned-tikzcode"
34
 
35
+ lora_config = LoraConfig(
36
+ r=16, # Rank of the decomposition, typically a small number (e.g., 8, 16)
37
+ lora_alpha=32, # Scaling factor for LoRA parameters
38
+ target_modules=["q_proj", "v_proj"], # Apply LoRA to specific model layers (e.g., Q, V projections in attention layers)
39
+ lora_dropout=0.1, # Dropout for LoRA layers
40
+ bias="none" # LoRA doesn’t update biases by default; change to "all" if needed
41
+ )
42
+
43
  # Load the model and processor
44
  model = MllamaForConditionalGeneration.from_pretrained(
45
  model_id,
46
  torch_dtype=torch.bfloat16,
47
  device_map="auto",
48
  )
49
+
50
+ model = get_peft_model(model, lora_config)
51
+
52
  processor = AutoProcessor.from_pretrained(model_id)
53
  ```