Upload finetune_soc.py with huggingface_hub

finetune_soc.py

@@ -6,6 +6,7 @@
 #     "peft>=0.7.0",
 #     "transformers>=4.36.0",
 #     "accelerate>=0.24.0",
+#     "bitsandbytes>=0.41.0",
 #     "datasets>=2.0.0",
 #     "trackio",
 # ]
@@ -22,7 +23,7 @@ Loss is computed on ALL tokens so the model learns every participant's voice.
 import trackio
 from datasets import load_dataset
 from peft import LoraConfig
-from transformers import AutoTokenizer
+from transformers import AutoTokenizer, BitsAndBytesConfig
 from trl import SFTConfig, SFTTrainer
 
 MODEL_ID = "swiss-ai/Apertus-8B-2509"
@@ -136,9 +137,17 @@ config = SFTConfig(
     run_name="apertus-8b-soc-v1",
 )
 
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype="bfloat16",
+    bnb_4bit_use_double_quant=True,
+)
+
 print("Initializing trainer...")
 trainer = SFTTrainer(
     model=MODEL_ID,
+    model_init_kwargs={"quantization_config": bnb_config, "device_map": "auto"},
     train_dataset=train_dataset,
     eval_dataset=eval_dataset,
     peft_config=peft_config,