float8

Float8 training provides substantial speedups for models where GEMMs are large enough that the FP8 tensorcore speedup outweighs dynamic quantization overhead.

Skills
AI research
0

What it does and how to use it

A skill that teaches Claude Code to do ai research tasks well. It is packaged knowledge the model uses whenever it is relevant.

Install the skill (command above).
In Claude Code, describe what you need - the skill turns on by itself when the context fits.
No commands to memorize: you talk in plain language and the skill does the rest.

Install (Claude Code)

mkdir -p .claude/skills/float8 && curl -fsSL https://doutorai.pt/data/lib/skills__float8.md -o .claude/skills/float8/SKILL.md

Paste this command into the terminal, in your project folder, and the component installs automatically.

Content

# Float8 Training in TorchTitan

Float8 training provides substantial speedups for models where GEMMs are large enough that the FP8 tensorcore speedup outweighs dynamic quantization overhead.

## Hardware Requirements

- NVIDIA H100 or newer GPUs (FP8 Tensor Cores)
- Blackwell GPUs for MXFP8 training

## Installation

```bash
USE_CPP=0 pip install git+https://github.com/pytorch/ao.git
```

## Usage: Tensorwise Scaling

Standard Float8 with tensorwise dynamic scaling:

```bash
CONFIG_FILE="./torchtitan/models/llama3/train_configs/llama3_8b.toml" ./run_train.sh \
  --model.converters="quantize.l

- 🔒 Full content on the Pro plan (€7.99/month) -

Need help with Claude Code?

Ask Dr AI - it explains things simply.

Need a tailor-made solution?

Not sure where to start with AI in your business? Get a free assessment and a tailored plan - no commitment.

Save hours every week with AI, no hassle

Doutor AI gives you 1,859 ready-to-copy AI prompts, by industry and in plain English - plus a complete technical library for those who want to go further.

See plans