menouar commited on
Commit
09bdd6c
1 Parent(s): 611507d

Use FalconForCausalLM for falcon finetuning instead of AutoModelForCausalLM

Browse files
Files changed (1) hide show
  1. utils/notebook_generator.py +7 -3
utils/notebook_generator.py CHANGED
@@ -2,7 +2,7 @@ from typing import Optional
2
 
3
  import nbformat as nbf
4
 
5
- from utils import FTDataSet
6
 
7
 
8
  def create_install_libraries_cells(cells: list):
@@ -130,9 +130,13 @@ def create_model_cells(cells: list, model_id: str, version: str, flash_attention
130
  if pad_value is None:
131
  pad_value_str = ""
132
 
 
 
 
 
133
  code = f"""
134
  import torch
135
- from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
136
  from trl import setup_chat_format
137
 
138
  # Hugging Face model id
@@ -145,7 +149,7 @@ bnb_config = BitsAndBytesConfig(
145
  )
146
 
147
  # Load model and tokenizer
148
- model = AutoModelForCausalLM.from_pretrained(
149
  model_id,
150
  device_map="auto",
151
  trust_remote_code=True,
 
2
 
3
  import nbformat as nbf
4
 
5
+ from utils import FTDataSet, falcon
6
 
7
 
8
  def create_install_libraries_cells(cells: list):
 
130
  if pad_value is None:
131
  pad_value_str = ""
132
 
133
+ auto_model_import = "AutoModelForCausalLM"
134
+ if model_id == falcon.name:
135
+ auto_model_import = "FalconForCausalLM"
136
+
137
  code = f"""
138
  import torch
139
+ from transformers import AutoTokenizer, {auto_model_import}, BitsAndBytesConfig
140
  from trl import setup_chat_format
141
 
142
  # Hugging Face model id
 
149
  )
150
 
151
  # Load model and tokenizer
152
+ model = {auto_model_import}.from_pretrained(
153
  model_id,
154
  device_map="auto",
155
  trust_remote_code=True,