SmolHub
A lightweight and efficient package for training language models using Low-Rank Adaptation (LoRA). Designed for easy experimentation and research with minimal boilerplate.
Features
- 🚀 Multiple Training Paradigms:
- Supervised Fine-tuning (SFT)
- Pretraining
- Preference Alignment (RLHF-style training)
- 📦 Easy Integration with Hugging Face models
- ⚡ Efficient Training with LoRA
- 📊 WandB Integration for experiment tracking
- 🔄 Automatic Dataset Handling
Installation
pip install smolhub
Quickstart
from smolhub.scripts.finetune import SFTTrainer
from smolhub.scripts.lora import LoRAModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
# Load model and tokenizer
model_id = "openai-community/gpt2"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)
# Setup LoRA
lora_model = LoRAModel(model)
# Train with minimal setup
trainer = SFTTrainer(
model=lora_model,
dataset_path="your_dataset", # HF dataset or local file
tokenizer=tokenizer
)
trainer.train()
Configuration
SmolHub uses a YAML configuration file for experiment settings. A default config is created in your project directory:
project:
name: SFTrainer
author: Yuvraj Singh
version: 1.0
LoRA:
rank: 4
alpha: 8
Preference:
beta: 0.1
Dataset:
use_hf_dataset: True
dataset_path: trl-lib/ultrafeedback_binarized
max_length: 512
batch_size: 16
num_workers: 4
shuffle: True
drop_last: True
pin_memory: True
persistent_workers: True
type: "classification" #TODO Add Chat style and Instruction
huggingface:
hf_token: "..."
Training:
type: 'preference'
Model:
epochs: 1
eval_iters: 10
eval_steps: 0
save_model_path: "saved_model"
saved_model_name: 'model.pt'
MAP:
use_bfloat16: False
use_float16: False
Optimizations:
use_compile: False
wandb:
project_name: "SFTrainer"
Training Modes
Supervised Fine-tuning
trainer = SFTTrainer(model, dataset_path="classification_dataset")
Pretraining
from smolhub.scripts.pretrain import PreTrainer
trainer = PreTrainer(model, dataset_path="text_corpus")
Preference Alignment
from smolhub.scripts.align import PreferenceTrainer
trainer = PreferenceTrainer(model, dataset_path="preference_pairs")
Documentation
For detailed documentation and examples, visit our documentation.
Contributing
Contributions are welcome! Please feel free to submit a Pull Request.
License
MIT License
1 maintainer
Metadata
1 maintainer
Release files for smolhub 0.6.8
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | |
|---|---|---|---|---|
| smolhub-0.6.8.tar.gz | 15.0 kB | Without provenance | Details |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| smolhub-0.6.8-py3-none-any.whl | Python 3 | none | any | Details |
Total release size: 35.6 kB
Release files / smolhub-0.6.8.tar.gz
| Download URL | smolhub-0.6.8.tar.gz |
|---|---|
| Size | 15.0 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
17c75624c6ff9f97cb0bfbc338d1c34a0092c2c13041c9acab8508f7659a019e
|
|
BLAKE2b-256 checksum How to use checksums |
7b3d2b59841b2030ac99d38ebd125e494eb8e072dffa6c946abfaab8c928a6b9
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.1.0 CPython/3.11.11
|
Release files / smolhub-0.6.8-py3-none-any.whl
| Download URL | smolhub-0.6.8-py3-none-any.whl |
|---|---|
| Size | 20.6 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
6b2cc4e48ed85c81af21807b2ead65dd5a461b028b2040c541b4c85087693d91
|
|
BLAKE2b-256 checksum How to use checksums |
1c7cf54ee87c9629f5c77c4b99f7fe036304fc6b69b152085066d5c732c2dacf
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.1.0 CPython/3.11.11
|
Security for smolhub 0.6.8
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.
