brsx 0.9.4

AutoForge — automatic model building, training, and fine-tuning that calibrates to your hardware. Fixed mamba bug and fixed critial NaN gradients in the maba scan under amp

brsx

AutoForge — build, train, and fine-tune models that calibrate themselves to your hardware.

brsx probes your GPU/CPU by actually running a few steps, picks a model size that fits your memory and speed target, and trains it. No config wrestling — pick a mode and go.

Install

pip install brsx

Optional extras (installed only if you need them):

pip install brsx[finetune]   # HuggingFace fine-tuning (transformers, peft, accelerate)
pip install brsx[qlora]      # 4-bit QLoRA (bitsandbytes, Linux/CUDA)
pip install brsx[tokenizer]  # fast BPE for the MTP branch
pip install brsx[all]        # everything

Usage

from brsx import automodel

automodel.run()

You'll get a menu:

1) Transformer training        — train a standard Transformer from scratch
2) MTP Transformer training    — Multi-Token Prediction heads (DeepSeek-V3 style)
3) HuggingFace model fine-tune — fine-tune a HF model (local path or hub id)
4) brsx (.pt) model fine-tune  — fine-tune an existing brsx model
5) fine-tune a mtp model
6) find right config for your pc
7) dataset editor for json, jsonl, parquet...
8) convert brsx model(.pt) to safetensors
9) A training system that can continue halfway with a special brsx technique without an optimizer load
10)Hybrid model training (transformer, mamba, gru, cnn all in one models.)
11)hybrid fine-tune
12) chat with a .safetensors brsx model or standard hf model(chat.py is reading the config, it will run every brsx and hf model.)

Pick one and follow the prompts — everything (mode, learning rate, steps, checkpointing, data source) is asked interactively.

License

Apache-2.0. See LICENSE.