Update README.md (Fix)

README.md

@@ -15,7 +15,7 @@ base_model: HuggingFaceTB/SmolLM3-3B-Base
 # Model Card for SmolLM3-3B-Instruct-Anime
 
 This model is a fine-tuned version of [HuggingFaceTB/SmolLM3-3B-Base](https://huggingface.co/HuggingFaceTB/SmolLM3-3B-Base).
-It has been trained using [zerofata/Instruct-Anime](https://huggingface.co/datasets/zerofata/Instruct-Anime).
+It was trained using [zerofata/Instruct-Anime](https://huggingface.co/datasets/zerofata/Instruct-Anime).
 
 ## Quick start
 
@@ -24,11 +24,11 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
 from peft import PeftModel
 
-# Define paths
+# Define the paths
 base_model_path = "./SmolLM3-3B-Base/"
 adapter_path = "./SmolLM3-3B-Instruct-Anime/"
 
-# Load the base model and tokenizer in bf16
+# Load the base model and the tokenizer in bf16
 print("Loading base model and tokenizer...")
 base_model = AutoModelForCausalLM.from_pretrained(
     base_model_path,
@@ -82,11 +82,11 @@ dataset_path = "./Instruct-Anime/instruct_dataset.jsonl"
 output_dir = "./SmolLM3-3B-Instruct-Anime"
 project_name = "smollm3-sft-anime"
 
-# --- 1. Initialize Tracking ---
+# --- 1. Initialize Trackio ---
 trackio.init(project=project_name)
 
-# --- 2. Load Model and Tokenizer ---
-print("Loading model and tokenizer...")
+# --- 2. Load the model and the tokenizer ---
+print("Loading the model and the tokenizer...")
 model = AutoModelForCausalLM.from_pretrained(
     model_name,
     device_map="auto",
@@ -101,20 +101,20 @@ if tokenizer.pad_token is None:
     tokenizer.pad_token = tokenizer.eos_token
     model.config.pad_token_id = model.config.eos_token_id
 
-# Load and set chat template from the jinja file
+# Load and set the chat template from the jinja file
 with open("chat_template.jinja", "r") as f:
     chat_template = f.read()
 tokenizer.chat_template = chat_template
-print("Chat template loaded from chat_template.jinja and set on the tokenizer.")
+print("The chat template has been loaded from chat_template.jinja and set on the tokenizer.")
 
 
-# --- Enable Gradient Checkpointing ---
+# --- Enable gradient checkpointing ---
 print("Enabling Gradient Checkpointing...")
 model.gradient_checkpointing_enable()
 
 
-# --- 3. Load and Process Dataset ---
-print("Loading and processing dataset...")
+# --- 3. Load and process the dataset ---
+print("Loading and processing the dataset...")
 dataset = load_dataset("json", data_files=dataset_path, split="train")
 
 def formatting_prompts_func(example):
@@ -128,8 +128,8 @@ dataset = dataset.map(formatting_prompts_func, remove_columns=["messages", "sour
 print(f"Dataset loaded and formatted with {len(dataset)} examples.")
 
 
-# --- 4. Configure LoRA ---
-print("Configuring LoRA...")
+# --- 4. Configure the LoRA ---
+print("Configuring the LoRA...")
 peft_config = LoraConfig(
     r=8,
     lora_alpha=16,
@@ -139,7 +139,7 @@ peft_config = LoraConfig(
     task_type="CAUSAL_LM",
 )
 
-# --- 5. Configure Training ---
+# --- 5. Configure training ---
 # Balanced learning rate and batch size for a GPU with ~24GB VRAM
 print("Configuring training arguments...")
 training_args = SFTConfig(
@@ -162,7 +162,7 @@ training_args = SFTConfig(
     greater_is_better=false
 )
 
-# --- 6. Create and Run Trainer ---
+# --- 6. Create and run the trainer ---
 print("Creating SFTTrainer...")
 trainer = SFTTrainer(
     model=model,
@@ -176,10 +176,10 @@ print("Starting training...")
 trainer.train() #resume_from_checkpoint=True
 
 # --- 7. Save the final adapter ---
-print("Training finished. Saving adapter.")
+print("Training has finished. Saving the adapter.")
 trainer.save_model(output_dir)
 
-print(f"LoRA adapter saved to {output_dir}")
+print(f"The LoRA adapter saved to {output_dir}")
 trackio.finish()
 ```