Benchmark result
unsloth/gemma-4-31B-it-qat-GGUF:UD-Q4_K_XL on AMD Radeon AI PRO R9700 — 53.7 tok/s
Measured with llama.cpp b11146-7fe450e19 on October 13, 2026.
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
2nd fastest of 5 runs of this model on AMD Radeon AI PRO R9700 · median 53.7 tok/s.
sort
- 154.0tok/s69.6llama.cpp · Q4_K
- 253.7tok/s73.5llama.cpp · Q4_Kthis run
- 353.7tok/s78.5llama.cpp · Q4_K
- 453.7tok/s70.8llama.cpp · Q4_K
- 553.4tok/s67.0llama.cpp · Q4_K
unsloth/gemma-4-31B-it-qat-GGUF:UD-Q4_K_XL on other hardware
Reproduce this run
toolllama.cpp b11146-7fe450e19
modelunsloth/gemma-4-31B-it-qat-GGUF:UD-Q4_K_XL (Q4_K)
context80,000
thinkingoff
samplingtemperature 1 · top_p 0.949999988079071 · top_k 64 · min_p 0.05000000074505806 · presence_penalty 0 · repeat_penalty 1
clientv0.4.92+102
Set those in llama.cpp, then:
llm-benchmark benchmark --model "unsloth/gemma-4-31B-it-qat-GGUF:UD-Q4_K_XL" --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.