🧠 Anime-Gen-Llama-2-7B

Anime-Gen-Llama-2-7B is a LoRA fine-tuned version of meta-llama/Llama-2-7b-hf, trained on a custom anime/manga-style dataset to generate structured short stories and panel descriptions from prompts. This model was trained using the PEFT library and bitsandbytes for efficient finetuning.

Model Details

Model Description

Developer: Vignesh Ramaswamy Balasundaram
Finetuned From: meta-llama/Llama-2-7b-hf
Language: English
License: Meta LLaMA 2 community license
Task Type: Causal Language Modeling
LoRA Adapter: Yes (via peft)
Model Size: 7B parameters (base model)

Model Sources

Repository: https://huggingface.co/vignesh0007/Anime-Gen-Llama-2-7B

Uses

Direct Use

Text-to-Anime panel generation
Short manga-style storytelling
Prompt-driven narrative generation

Out-of-Scope Use

Legal, financial, or medical decision-making
Real-time conversation agents
General-purpose dialogue

Bias, Risks, and Limitations

Limitations

Trained on a small, domain-specific dataset (anime-style stories)
May hallucinate character names or plots
Not guaranteed to follow strict story structure

Recommendations

Users should validate outputs for correctness and coherence before using in production.
Consider using as a creative writing aid rather than factual generation.

How to Get Started

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

adapter_id = "vignesh0007/Anime-Gen-Llama-2-7B"

config = PeftConfig.from_pretrained(adapter_id)

base_model = AutoModelForCausalLM.from_pretrained(
    config.base_model_name_or_path,
    device_map="auto",
    trust_remote_code=True
)

model = PeftModel.from_pretrained(base_model, adapter_id)
tokenizer = AutoTokenizer.from_pretrained(adapter_id)

prompt = "Title: The Final Duel\nCharacters: Yuki, Daichi\nPanel 1:"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=100, temperature=0.8, top_p=0.95, do_sample=True)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))