Upgrade transformers version and handle device assignment in llama trainer

requirements-no-gpu-uv.txt

@@ -132,7 +132,7 @@ httpcore==1.0.7
     # via httpx
 httpx==0.28.1
     # via fschat
-huggingface-hub==0.28.0
+huggingface-hub==0.30.2
     # via
     #   accelerate
     #   datasets
@@ -450,7 +450,7 @@ tqdm==4.66.5
     #   peft
     #   sentence-transformers
     #   transformers
-transformers==4.50.0
+transformers==4.51.0
     # via
     #   -r requirements.in
     #   peft

requirements-uv.txt

@@ -132,7 +132,7 @@ httpcore==1.0.7
     # via httpx
 httpx==0.28.1
     # via fschat
-huggingface-hub==0.28.0
+huggingface-hub==0.30.2
     # via
     #   accelerate
     #   datasets
@@ -482,7 +482,7 @@ tqdm==4.67.1
     #   peft
     #   sentence-transformers
     #   transformers
-transformers==4.50.0
+transformers==4.51.0
     # via
     #   -r requirements.in
     #   peft

requirements.in

@@ -21,7 +21,7 @@ tiktoken
 torch==2.6.0
 torchaudio==2.6.0
 torchvision==0.21.0
-transformers==4.50.0
+transformers==4.51.0
 peft
 watchfiles
 wandb==0.19.8

transformerlab/plugins/llama_trainer/index.json

@@ -4,7 +4,7 @@
   "description": "A training script adapted from https://www.philschmid.de/instruction-tune-llama-2 for training Llama2 using PeFT",
   "plugin-format": "python",
   "type": "trainer",
-  "version": "1.0.18",
+  "version": "1.0.19",
   "model_architectures": ["LlamaForCausalLM", "Qwen2ForCausalLM"],
   "git": "",
   "url": "",

transformerlab/plugins/llama_trainer/main.py

@@ -3,6 +3,9 @@ import os
 from random import randrange
 
 import torch
+if torch.cuda.is_available():
+    os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+    
 from jinja2 import Environment
 from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training, PeftModel
 from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, AutoConfig
@@ -16,9 +19,6 @@ use_flash_attention = False
 # Initialize Jinja environment
 jinja_environment = Environment()
 
-if torch.cuda.is_available():
-    os.environ["CUDA_VISIBLE_DEVICES"] = "0"
-
 
 @tlab_trainer.job_wrapper()
 def train_model():