vincentwi's picture
Add model card
f5eeeb7 verified
metadata
license: llama3.1
base_model: meta-llama/Llama-3.1-8B-Instruct
tags:
  - ascii-art
  - fine-tuned
  - llama
  - art-generation

haizelabs/sft-svgeez-blocks-20251101T005904Z-checkpoint-6500

This is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct specialized for generating ASCII art.

Model 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

Usage

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-blocks-20251101T005904Z-checkpoint-6500")

# 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

  • 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.