Upload folder using huggingface_hub

app.py

@@ -1,11 +1,8 @@
 import gradio as gr
 import os
 import torch
-from threading import Thread
-import time
 
-# Training status
-training_status = {"running": False, "log": "", "progress": 0}
+training_status = {"running": False, "log": ""}
 
 def run_training(
     base_model: str,
@@ -41,7 +38,7 @@ def run_training(
         from datasets import load_dataset
         from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig, TrainingArguments
         from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
-        from trl import SFTTrainer
+        from trl import SFTTrainer, SFTConfig
         
         progress(0.15, desc="Loading tokenizer...")
         log(f"[3/6] Loading tokenizer: {base_model}")
@@ -96,7 +93,8 @@ def run_training(
         progress(0.4, desc="Setting up trainer...")
         log(f"[6/6] Starting training: {epochs} epochs, batch={batch_size}, lr={learning_rate}")
         
-        training_args = TrainingArguments(
+        # Use SFTConfig instead of TrainingArguments for newer TRL
+        sft_config = SFTConfig(
             output_dir="./outputs",
             num_train_epochs=epochs,
             per_device_train_batch_size=batch_size,
@@ -111,15 +109,15 @@ def run_training(
             push_to_hub=True,
             hub_model_id=output_repo,
             hub_token=os.environ.get("HF_TOKEN"),
+            max_seq_length=4096,
+            dataset_text_field="text",
         )
         
         trainer = SFTTrainer(
             model=model,
-            args=training_args,
+            args=sft_config,
             train_dataset=dataset,
-            tokenizer=tokenizer,
-            max_seq_length=4096,
-            dataset_text_field="text",
+            processing_class=tokenizer,
         )
         
         log("\n" + "=" * 50)
@@ -154,6 +152,7 @@ with gr.Blocks(title="Agent Zero Trainer") as demo:
     **Intuition Labs** • terminals.tech
     
     Fine-tune models for coherent multi-context orchestration.
+    Running on L40S GPU (48GB VRAM) - $1.80/hr
     """)
     
     with gr.Row():