detect logged in user

app.py

@@ -19,6 +19,15 @@ logger = logging.getLogger(__name__)
 log_contents = log_stream.getvalue()
 print(log_contents)
 logger.debug('This is a debug message')
+
+hf_user = None
+hfApi = HfApi()
+try:
+    hf_user = hfApi.whoami()
+except Exception as e:
+    hf_user = "not logged in"
+
+
 # Dropdown options
 model_options = [
     "unsloth/mistral-7b-v0.3-bnb-4bit",      # New Mistral v3 2x faster!
@@ -53,7 +62,7 @@ class PrinterCallback(TrainerCallback):
     def on_step_end(self, args, state, control, **kwargs):
         if state.is_local_process_zero:
             self.step = state.global_step
-            self.progress.update(self.step)
+            self.progress(self.step/60, desc=f"Training {self.step}/60")
             #print("**Step ", state.global_step)
 
 
@@ -106,55 +115,6 @@ def inference(prompt, input_text):
     result = tokenizer.batch_decode(outputs)
     return result[0], gr.update(visible=True, interactive=True)
 
-
-async def train_model(model_name: str, lora_r: int, lora_alpha: int, lora_dropout: float, per_device_train_batch_size: int, warmup_steps: int, max_steps: int,
-        gradient_accumulation_steps: int, logging_steps: int, log_to_tensorboard: bool, optim, learning_rate, weight_decay, lr_scheduler_type, seed: int, output_dir, progress= gr.Progress()):
-    global model, tokenizer
-    print(f"$$$ Training model {model_name} with {lora_r} R, {lora_alpha} alpha, {lora_dropout} dropout, {per_device_train_batch_size} per device train batch size, {warmup_steps} warmup steps, {max_steps} max steps, {gradient_accumulation_steps} gradient accumulation steps, {logging_steps} logging steps, {log_to_tensorboard} log to tensorboard, {optim} optimizer, {learning_rate} learning rate, {weight_decay} weight decay, {lr_scheduler_type} lr scheduler type, {seed} seed, {output_dir} output dir")
-    iseed = seed
-    model = FastLanguageModel.get_peft_model(
-        model,
-        r = lora_r,
-        target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
-                        "gate_proj", "up_proj", "down_proj",],
-        lora_alpha = lora_alpha,
-        lora_dropout = lora_dropout,
-        bias = "none",
-        use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
-        random_state=iseed,
-        use_rslora = False,  # We support rank stabilized LoRA
-        loftq_config = None, # And LoftQ
-    )
-
-    trainer = SFTTrainer(
-        model = model,
-        tokenizer = tokenizer,
-        train_dataset = dataset,
-        dataset_text_field = "text",
-        max_seq_length = max_seq_length,
-        dataset_num_proc = 2,
-        packing = False, # Can make training 5x faster for short sequences.
-        callbacks = [PrinterCallback(progress)],
-        args = TrainingArguments(
-            per_device_train_batch_size = per_device_train_batch_size,
-            gradient_accumulation_steps = gradient_accumulation_steps,
-            warmup_steps = warmup_steps,
-            max_steps = 60, # Set num_train_epochs = 1 for full training runs
-            learning_rate = learning_rate,
-            fp16 = not is_bfloat16_supported(),
-            bf16 = is_bfloat16_supported(),
-            logging_steps = logging_steps,
-            optim = "adamw_8bit",
-            weight_decay = weight_decay,
-            lr_scheduler_type = "linear",
-            seed = iseed,
-            report_to="tensorboard" if log_to_tensorboard else None,
-            output_dir = output_dir
-        ),
-    )
-    trainer.train()
-    return "Model training",gr.update(visible=True, interactive=False), gr.update(visible=True, interactive=True), gr.update(interactive=True)
-
 def save_model(model_name, hub_model_name, hub_token, gguf_16bit, gguf_8bit, gguf_4bit, gguf_custom, gguf_custom_value, merge_16bit, merge_4bit, just_lora, push_to_hub):
     global model, tokenizer
     if gguf_custom:
@@ -190,7 +150,7 @@ with gr.Blocks(title="Unsloth fine-tuning") as demo:
     with gr.Column():
         gr.Image("unsloth.png", width="300px", interactive=False, show_download_button=False, show_label=False)
     with gr.Column():
-        gr.Markdown(f"**GPU Information:** {gpu_stats.name} ({max_memory} GB)\n\n[Unsloth Docs](http://docs.unsloth.com/)\n\n[Unsloth GitHub](https://github.com/unslothai/unsloth)") 
+        gr.Markdown(f"**User:** {hf_user}\n\n**GPU Information:** {gpu_stats.name} ({max_memory} GB)\n\n[Unsloth Docs](http://docs.unsloth.com/)\n\n[Unsloth GitHub](https://github.com/unslothai/unsloth)") 
     with gr.Tab("Base Model Parameters"):
 
         with gr.Row():
@@ -265,6 +225,59 @@ with gr.Blocks(title="Unsloth fine-tuning") as demo:
 
         train_output = gr.Textbox(label="Training Status", value="Model not trained", interactive=False)
         train_btn = gr.Button("Train", visible=True)
+
+        def train_model(model_name: str, lora_r: int, lora_alpha: int, lora_dropout: float, per_device_train_batch_size: int, warmup_steps: int, max_steps: int,
+                gradient_accumulation_steps: int, logging_steps: int, log_to_tensorboard: bool, optim, learning_rate, weight_decay, lr_scheduler_type, seed: int, output_dir, progress= gr.Progress()):
+            global model, tokenizer
+            print(f"$$$ Training model {model_name} with {lora_r} R, {lora_alpha} alpha, {lora_dropout} dropout, {per_device_train_batch_size} per device train batch size, {warmup_steps} warmup steps, {max_steps} max steps, {gradient_accumulation_steps} gradient accumulation steps, {logging_steps} logging steps, {log_to_tensorboard} log to tensorboard, {optim} optimizer, {learning_rate} learning rate, {weight_decay} weight decay, {lr_scheduler_type} lr scheduler type, {seed} seed, {output_dir} output dir")
+            iseed = seed
+            model = FastLanguageModel.get_peft_model(
+                model,
+                r = lora_r,
+                target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
+                                "gate_proj", "up_proj", "down_proj",],
+                lora_alpha = lora_alpha,
+                lora_dropout = lora_dropout,
+                bias = "none",
+                use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
+                random_state=iseed,
+                use_rslora = False,  # We support rank stabilized LoRA
+                loftq_config = None, # And LoftQ
+            )
+            progress(0.0, desc="Loading Trainer")
+            time.sleep(1)
+            trainer = SFTTrainer(
+                model = model,
+                tokenizer = tokenizer,
+                train_dataset = dataset,
+                dataset_text_field = "text",
+                max_seq_length = max_seq_length,
+                dataset_num_proc = 2,
+                packing = False, # Can make training 5x faster for short sequences.
+                callbacks = [PrinterCallback(progress)],
+                args = TrainingArguments(
+                    per_device_train_batch_size = per_device_train_batch_size,
+                    gradient_accumulation_steps = gradient_accumulation_steps,
+                    warmup_steps = warmup_steps,
+                    max_steps = 60, # Set num_train_epochs = 1 for full training runs
+                    learning_rate = learning_rate,
+                    fp16 = not is_bfloat16_supported(),
+                    bf16 = is_bfloat16_supported(),
+                    logging_steps = logging_steps,
+                    optim = "adamw_8bit",
+                    weight_decay = weight_decay,
+                    lr_scheduler_type = "linear",
+                    seed = iseed,
+                    report_to="tensorboard" if log_to_tensorboard else None,
+                    output_dir = output_dir
+                ),
+            )
+            trainer.train()
+            progress(1, desc="Training completed")
+            time.sleep(1)
+            return "Model trained 100%",gr.update(visible=True, interactive=False), gr.update(visible=True, interactive=True), gr.update(interactive=True)
+
+
         train_btn.click(train_model, inputs=[model_name, lora_r, lora_alpha, lora_dropout, per_device_train_batch_size, warmup_steps, max_steps, gradient_accumulation_steps, logging_steps, log_to_tensorboard, optim, learning_rate, weight_decay, lr_scheduler_type, seed, output_dir], outputs=[train_output, train_btn])
 
     with gr.Tab("Save & Push Options"):