fp8 quant made via llm-compressor. compressed via ztsd due to slow internet.
Linux
Install zstd:
# Ubuntu/Debian
sudo apt install zstd
# RHEL/CentOS/Fedora
sudo dnf install zstd
Extract (multi-threaded):
# Single command (if tar supports zstd)
tar -I 'zstd -T0' -xf Austral-FP8.tar.zst
# Two-step method
zstd -d -T0 Austral-FP8.tar.zst
tar -xf Austral-FP8.tar
macOS
Install zstd:
brew install zstd
Extract (multi-threaded):
# Single command
tar -I 'zstd -T0' -xf Austral-FP8.tar.zst
# Two-step method
zstd -d -T0 Austral-FP8.tar.zst
tar -xf Austral-FP8.tar
Windows
Install zstd:
- Download from Facebook/zstd releases
- Or use Chocolatey:
choco install zstandard
- Or use WSL with Linux instructions
Extract (multi-threaded):
PowerShell/CMD:
zstd.exe -d -T0 Austral-FP8.tar.zst
tar -xf Austral-FP8.tar
WSL:
zstd -d -T0 Austral-FP8.tar.zst
tar -xf Austral-FP8.tar
Options Explained
-d
: Decompress-T0
: Use all available CPU cores-T4
: Use 4 threads (specify number)--rm
: Remove source file after extraction
Notes
- Maximum compression means slower decompression but better file size
- Multi-threading (
-T0
) significantly speeds up extraction on multi-core systems
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
๐
Ask for provider support
Model tree for NewEden/Austral-70B-FP8
Base model
meta-llama/Llama-3.1-70B
Finetuned
meta-llama/Llama-3.3-70B-Instruct
Finetuned
Sao10K/Llama-3.3-70B-Vulpecula-r1
Finetuned
Delta-Vector/Austral-70B-Preview