Update README.md

README.md

@@ -9,81 +9,27 @@ This is a custom model for text generation.
 
 ## Model Details
 
-- `model_type`: Sparkoo
+- `model_type`: GPT2*
 
-## Example usage
+## GPT2
 
-```python
-import torch
-from transformers import GPT2LMHeadModel, GPT2Tokenizer
+This model is **NOT A FINETUNE!!**. It uses the GPT2 architecture but it doesnt finetune it.
 
-def generate_text(prompt, model_name, max_length=100, num_return_sequences=1):
-    """
-    Generate text using the Sparkoo/KateAI model from Hugging Face Hub.
-    
-    Args:
-        prompt (str): The input text to start generation from
-        model_name (str): Name of the model on Hugging Face Hub
-        max_length (int): Maximum length of generated text
-        num_return_sequences (int): Number of different sequences to generate
-    """
-    # Load model and tokenizer
-    print(f"Loading model from {model_name}...")
-    model = GPT2LMHeadModel.from_pretrained(model_name)
-    tokenizer = GPT2Tokenizer.from_pretrained(model_name)  # Use original GPT2 tokenizer
-    tokenizer.pad_token = tokenizer.eos_token
-    
-    # Move model to GPU if available
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    model = model.to(device)
-    model.eval()
-    
-    # Encode the input prompt
-    encoded_prompt = tokenizer(prompt, return_tensors="pt", padding=True).to(device)
-    
-    # Generate text
-    print("\nGenerating text...")
-    with torch.no_grad():
-        output_sequences = model.generate(
-            input_ids=encoded_prompt["input_ids"],
-            attention_mask=encoded_prompt["attention_mask"],
-            max_length=max_length,
-            temperature=0.7,
-            top_k=50,
-            top_p=0.95,
-            do_sample=True,
-            num_return_sequences=num_return_sequences,
-            pad_token_id=tokenizer.eos_token_id,
-            eos_token_id=tokenizer.eos_token_id
-        )
-    
-    # Decode and print the generated text
-    for idx, sequence in enumerate(output_sequences):
-        generated_text = tokenizer.decode(sequence, skip_special_tokens=True)
-        print(f"\nGenerated sequence {idx + 1}:")
-        print(f"{generated_text}")
-        print("-" * 50)
+```python
+# Model configuration for a smaller GPT-2 style model
+    config = GPT2Config(
+        vocab_size=50257,  # Standard GPT-2 vocabulary size
+        n_positions=512,   # Maximum sequence length
+        n_ctx=512,        # Context window size
+        n_embd=512,       # Embedding dimension
+        n_layer=6,        # Number of transformer layers
+        n_head=8,         # Number of attention heads
+        bos_token_id=50256,
+        eos_token_id=50256,
+        pad_token_id=50256,
+        _name_or_path=""  # Empty to ensure no pretrained weights are loaded
+    )
 
-if __name__ == "__main__":
-    # Example prompts to test
-    prompts = [
-        "Once upon a time",
-        "The artificial intelligence",
-        "In the distant future",
-        "The scientist discovered"
-    ]
-    
-    model_name = "Sparkoo/KateAI50m"
-    
-    # Generate text for each prompt
-    for prompt in prompts:
-        print("\n" + "="*50)
-        print(f"Prompt: {prompt}")
-        print("="*50)
-        generate_text(
-            prompt=prompt,
-            model_name=model_name,
-            max_length=200,  # Adjust as needed
-            num_return_sequences=3
-        ) 
+    # Initialize model with random weights
+    model = GPT2LMHeadModel(config)
 ```