This model is a fine-tuned version of unsloth/Llama-3.2-3B, optimized for generating ASCII-style text art.
It was trained using Unsloth for efficient fine-tuning and Hugging Face TRL for supervised training.

Base model: unsloth/Llama-3.2-3B
Fine-tuned by: shalhamucha
License: Apache 2.0
Language: English
Frameworks: Transformers, TRL, Unsloth

Training Details

Method: LoRA (Low-Rank Adaptation)
Target modules: q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
Rank (r): 16
LoRA α (alpha): 16
LoRA dropout: 0
Optimizer: AdamW (8-bit, via bitsandbytes)
Precision: bf16/fp16 (depending on hardware support)
Context length: 2048 tokens
Training epochs: 2
Batch size (effective): 8 (2 × gradient accumulation steps of 4)

The model was trained on the shalhamucha/cats dataset.

Intended Use

Primary: Generating ASCII cat art 🐱
Secondary: Can be adapted for general ASCII-style outputs or text art generation tasks.
Not intended for: High-stakes decision-making, factual QA, or reasoning-heavy tasks.

Example Usage

python

from unsloth import FastLanguageModel
from transformers import TextStreamer

model, tokenizer = FastLanguageModel.from_pretrained(
    "shalhamucha/Llama-3.2-3B-ascii-cats-guide-lora",
    max_seq_length=2048,
    dtype=None,
    load_in_4bit=True
)

FastLanguageModel.for_inference(model)

inputs = tokenizer("", return_tensors="pt").to("cuda")
streamer = TextStreamer(tokenizer)

model.generate(**inputs, streamer=streamer, max_new_tokens=100)

Example output:

     /\_/\  
    ( o.o ) 
     > ^ <

Citation

@misc{shalhamucha2025asciicats,
  title = {Llama-3.2-3B ASCII Cats (LoRA Fine-tuned)},
  author = {Shalha Mucha Bhuyan},
  year = {2025},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/shalhamucha/Llama-3.2-3B-ascii-cats-guide-lora}},
}