Environment Specifications
- Repository: kyuz0/strix-halo-ds4-toolbox
- Hardware: AMD Ryzen AI Max "Strix Halo"
- Memory: 128GB vRAM (Configured with Unified Memory)
- Software Stack: Fedora 43, Kernel 6.19.12-200.fc43.x86_64
- Models: Multiple quants compared
Prefill Throughput (Tokens/s)
Prompt processing speed across expanding context lengths.
Generation Throughput (Tokens/s)
Token generation speed across expanding context lengths.
Raw Benchmark Data
Detailed breakdown of tokens/s across selected models.