metadata
license: other
license_name: haste
license_link: LICENSE
language:
- en
pipeline_tag: text-generation
tags:
- English
- Generation
- Small
Haste
Haste is a high-performance, lightweight language model designed for fast inference and efficient training. Released under the Haste License, it is optimized for small-to-medium scale natural language processing tasks, making it suitable for both research and practical applications.
Key Features
- Parameters: 150M
- Fast Inference: Optimized for low-latency text generation
- Lightweight: Small footprint allows deployment on consumer-grade hardware
- Versatile: Supports text generation, completion, and fine-tuning tasks
- Easy to Use: Compatible with Hugging Face Transformers ecosystem
Model Details
| Feature | Details |
|---|---|
| Model Name | Haste |
| License | Haste License |
| Parameter Count | 150M |
| Framework | Hugging Face Transformers |
| Precision | FP16 / mixed precision supported |
| Device Support | CPU / GPU / Multi-GPU |
Installation
pip install transformers torch
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model_name = "theguywhosucks/haste"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.float16)
input_text = "Hi"
inputs = tokenizer(input_text, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Training
Haste can be further fine-tuned using standard Hugging Face Trainer APIs. It is optimized for small datasets and limited GPU resources.
Contributing
I welcome contributions! Please follow the code of conduct and ensure all pull requests are well-documented and tested.
License
Haste is released under the Haste License. For full license terms, see the LICENSE file.