# Poros documentation

Fine-tune large models on small GPUs: 32B in ~11 GB of training peak, and 72B
end-to-end on a single 32 GB card.

Poros streams the frozen quantized base through the GPU one block at a time,
keeping only the LoRA adapters resident in VRAM. The streaming changes
nothing — loss, adapter weights, and optimizer state all come out bit-for-bit
identical to the same configuration trained fully resident.

## Getting started

- [Installation](install.md)
- [Quickstart](quickstart.md)
- [Train on your own data](poros-trainer.md)

## Usage

- [`poros.prepare` Auto API](auto.md)
- [Adapter support matrix](adapters.md)
- [HF drop-in patch (`patch_hf_model`)](api.md#hf-drop-in-patch)
- [PorosTrainer API](api.md#porostrainer)
- [PorosConfig reference](api.md#porosconfig)
- [CLI reference](api.md#cli-reference)
- [Benchmark guide](bench.md)

## Reference

- [Consumer GPUs guide (sizing, OOM playbook)](consumer-gpus.md)
- [Supported architectures](architecture-support.md)
- [Limitations and roadmap](limitations.md)
- [Validation artifacts (raw evidence behind every number)](validation/README.md)

## Key properties

- **Bitwise parity**: 0.00e+00 max diff on loss and adapter weights (NF4
  QLoRA, gated 0.00e+00 across Qwen2.5, Qwen3, Gemma3, and Gemma4 on the
  platforms in the paper's parity matrix, Gemma on RTX 5090; plus Qwen3.5/Qwen3.6
  validated on RTX 4090).
- **68–77% VRAM reduction** vs fully-resident QLoRA; 45–60% less than
  Unsloth (NF4, 32B–72B scale).
- **Throughput tradeoff**: hardware-dependent — about 1.6× slower per step on
  an RTX PRO 6000, up to ~2.7× on an A100, and ~3.5–4.3× on a B300 (where the
  resident baseline is so fast that PCIe streaming dominates the ratio).
  Memory for time.
- **Pure-python install**: `pip install poros-train` works without torch.
  ML dependencies are optional extras.