from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import torch

# Load the base model and tokenizer
base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")

# Load the fine-tuned adapter
model = PeftModel.from_pretrained(base_model, "haizelabs/sft-svgeez-fresh-20251028T015914Z-checkpoint-10000")

# Example usage
def generate_ascii_art(prompt):
    messages = [
        {"role": "system", "content": "You are an expert ASCII artist. Generate clean, artistic ASCII representations of the requested objects."},
        {"role": "user", "content": prompt}
    ]
    
    input_text = tokenizer.apply_chat_template(messages, tokenize=False)
    inputs = tokenizer(input_text, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=1024,
            do_sample=True,
            temperature=0.7,
            pad_token_id=tokenizer.eos_token_id
        )
    
    response = tokenizer.decode(outputs[0][len(inputs.input_ids[0]):], skip_special_tokens=True)
    return response

# Generate ASCII art
ascii_art = generate_ascii_art("Draw an ASCII image of a cat")
print(ascii_art)

Training Details

Base Model: meta-llama/Llama-3.1-8B-Instruct
Fine-tuning Method: Supervised Fine-Tuning (SFT) with LoRA
Dataset: ASCII Bench Haiku dataset
Purpose: Generate ASCII art from text descriptions

Training Steps: 672
Learning Rate: 5e-4
Batch Size: 12 (per device)
Gradient Accumulation Steps: 3
LoRA Rank: 128
LoRA Alpha: 256

Limitations

This model is fine-tuned specifically for ASCII art generation and may not perform well on other tasks. The quality of ASCII art generation depends on the complexity and clarity of the input prompt.