Benchmark result
ds-v4-mtp-q4km on AMD Radeon 680M — 8.9 tok/s
Measured with llama.cpp manual on November 14, 2026.
Support llm-bench.io
llm-bench.io is free. If it helped you, I'd appreciate your support to cover the infrastructure costs.
What the model built
Open full screen →The coding scenario asks for a playable game in a single HTML file. This is exactly what the model returned, unedited; the frame only adds a line that tells this page how tall it is.
How this run compares
1st fastest of 2 runs of this model on AMD Radeon 680M · median 8.9 tok/s.
sort
- 18.9tok/s74.2llama.cppthis run
- 28.8tok/s67.9llama.cpp
Reproduce this run
toolllama.cpp manual
modelds-v4-mtp-q4km
context8,192
think limit16,384 tokens
samplingtemperature 0.6000000238418579 · top_p 0.949999988079071 · top_k 20 · min_p 0 · presence_penalty 0 · repeat_penalty 1
clientv0.4.99+102
Set those in llama.cpp, then:
llm-benchmark benchmark --model "ds-v4-mtp-q4km" --tool "llama.cpp"The client prompts for context length and thinking mode, and for the KV cache dtype on Unsloth Studio. Sampling is left at the model default — the values above are what the tool reported using, not overrides the client sent.