models / collection
Every port, with its receipts.
Each model started as a GGUF on Hugging Face. Open one for every workload against llama.cpp ROCm on the same GGUF, its serving configuration, weights, gates and provenance.
loom llama.cpp, tok/snumbers: loom over llama.cpp
tok/s: loom over llama.cpp
gate pass · fail · no evidence
loom ÷ llama.cpp, same GGUF:
below 1.00
1.00–1.20
~1.40
1.80+
— not measured
Gates compare the port with llama.cpp on the same GGUF. A hollow square means the gate applies but no recorded run covers it yet, and a dot means it doesn't apply. Hover a column header for what it checks.
■ pass · ■ fail · □ no evidence · · n/a
■ feature supported · · not supported