Tech Tweet Generator Llama-3 (Fine-Tuned)
This model is a fine-tuned version of Llama-3.2-3B-Instruct designed to convert dense scientific and technical research paper abstracts into engaging, viral Twitter threads.
It was trained using LoRA (Low-Rank Adaptation) on the Apple MLX framework.
π Model Description
- Developed by: Meet Merchant
- Base Model:
mlx-community/Llama-3.2-3B-Instruct-4bit - Task: Summarization / Style Transfer (Research Paper Abstract -> Engaging Twitter Thread)
- Language: English
- Framework: MLX
π» How to Use
You can run this model locally on your Mac using mlx-lm.
Installation
pip install mlx-lm
Python Code
from mlx_lm import load, generate
# Load the model and your adapters
model, tokenizer = load(
"mlx-community/Llama-3.2-3B-Instruct-4bit",
adapter_path="meetmerchant/tech-tweet-generator-llama3"
)
abstract = """
[Paste Abstract Here]
"""
prompt = f"""
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
You are a viral science communicator.
<|eot_id|><|start_header_id|>user<|end_header_id|>
Title: Example Paper
Abstract:
{abstract}
<|eot_id|><|start_header_id|>assistant<|end_header_id|>
"""
response = generate(model, tokenizer, prompt=prompt, max_tokens=500)
print(response)
π Training Details
- Dataset: 50+ ArXiv papers from the field of AI.
- Ground Truth: Generated by GPT-4o-mini ("Teacher-Student" distillation).
- Training Config:
- LoRA Rank: 16
- Quantization: 4-bit
- Iterations: 200
π Evaluation
The model was evaluated using LLM-as-a-Judge (GPT-4o) against the base model.
- Win Rate: 66% (vs Base Model)
- Strengths: High engagement, emoji usage, accessible language.
- Weaknesses: Can occasionally hallucinate specific details if the abstract is too dense.
Built with β€οΈ on a MacBook Pro M3 using Apple MLX.
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
π
Ask for provider support
Model tree for meetmerchant/tech-tweet-generator-llama3
Base model
meta-llama/Llama-3.2-3B-Instruct
Finetuned
mlx-community/Llama-3.2-3B-Instruct
Quantized
mlx-community/Llama-3.2-3B-Instruct-4bit