Tessera: trellis-coded quantization of LLM weights onto hardware-native NVFP4, FP8 and BF16 tiles, continuous-rate on the FP8 and BF16 grids, with its own vLLM serving plugin
vLLM plugin for variable-rate NVFP4-CB / FP8-CB product-codebook LLM quantization served by dedicated CUDA decode and prefill kernels.
Mixed-precision quantization allocator for LLMs: every layer refracts into a different format based on its sensitivity.
Anthropic, PBC
Visionary sponsor
Bloomberg
Visionary sponsor
Hudson River Trading
Visionary sponsor
Meta
Visionary sponsor
NVIDIA
Visionary sponsor
Microsoft
Sustainability sponsor
Depot
Continuous Integration
AWS
Cloud computing and Security Sponsor
Datadog
Monitoring
Fastly
CDN
Google
Download Analytics
Sentry
Error logging
StatusPage
Status page