Lab / / failure
3.754× behind ggml, on weaker footing
SAM3 x64 encode measured 3.754× behind ggml and Qwen 6.18× behind llama.cpp. The report gives the ratios but not the hosts, thread counts, or build flags.
Negative results, Measurement methodology, BenchmarkingWeft- Machine
- UNKNOWN — not tied to a host
- Commit
- UNKNOWN
- SAM3 x64 encode vs ggml
- 3.754× behind
- Qwen vs llama.cpp, end to end
- 6.18× behind
Two more losses. SAM3 x64 encode sits 3.754× behind ggml, and Qwen sits 6.18× behind llama.cpp end to end.
They belong on the site precisely because they are losses. A body of work that only ever reports wins is not reporting.
Why this is MOSTLY_MATCHED and not MATCHED
The report gives the ratio but not the host, thread count, or build flags for either figure. Two end-to-end runs of the same model on different machines are not a matched comparison, and nothing in the document proves they were on the same one. The report also does not tie either figure to a host on the 8× RTX 5090 box used for the CUDA comparison.
| Comparison | Result |
|---|---|
| SAM3 x64 encode vs ggml | 3.754× behind ggml |
| Qwen, end to end vs llama.cpp | 6.18× behind llama.cpp |
**Environment.** UNKNOWN — the report does not state the host, thread count, or build flags for either figure, and does not tie either to a named machine.
**Methodology.** MOSTLY_MATCHED. Both figures are slowdown factors, so higher is worse for Weft. They are quoted because the comparison quality is MOSTLY rather than UNMATCHED; the missing host and flags are what stop it being matched.