Public KG record
ZYFA
Recorded overview
A set of FP8 and BF16 prefill attention kernels for AMD GPUs using Lean P and tw
An evidence-backed entity from the project knowledge graph; detailed tool metadata is still being added.
Recorded details
- Benchmarks: 2,672 TFLOP/s FP8 at 32K tokens on Gaussian benchmarks
- Capabilities: Lean P
- Capabilities: Two-wave ping-pong schedule
- License: Apache 2.0
- Supported types: text input/output
- Performance: 1.16-1.76x AITER's FP8 throughput on real activations
Evidence coverage
The KG currently records 11 known information dimensions and has 8 undocumented dimensions, linked to 2 public source articles. These counts describe record coverage, not model performance or verified fit.
Related public records
- Zyphra — releases
- DeepSeek-V2-Lite — evaluated on
- Mistral-7B — evaluated on
- Qwen3-8B — evaluated on
- Lean P — has feature