Benchmark result
prism-ml/Ternary-Bonsai-2-9B-gguf:PQ2_0 on NVIDIA GeForce RTX 4070 — 186.2 tok/s
Measured with llama.cpp manual on October 16, 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
8th fastest of 22 runs of this model on NVIDIA GeForce RTX 4070 · median 180.7 tok/s.
sort
- 1203.6tok/s64.8llama.cpp · Q2_0
- 2196.9tok/s58.2llama.cpp · Q2_0
- 3194.6tok/s52.4llama.cpp · Q2_0
- 4191.6tok/s56.2llama.cpp · Q2_0
- 5187.8tok/s59.5llama.cpp · Q2_0
- 6187.4tok/s60.1llama.cpp · Q2_0
- 7187.3tok/s52.7llama.cpp · Q2_0
- 8186.2tok/s61.4llama.cpp · Q2_0this run
- 9186.0tok/s55.2llama.cpp · Q2_0
- 10183.6tok/s57.4llama.cpp · Q2_0
- 11183.5tok/s50.8llama.cpp · Q2_0
- 12178.0tok/s58.0llama.cpp · Q2_0
- 13174.1tok/s52.4llama.cpp · Q2_0
- 14163.9tok/s65.2llama.cpp · Q2_0
- 15162.0tok/s49.0llama.cpp · Q2_0
- 16161.1tok/s51.8llama.cpp · Q2_0
- 17158.9tok/s63.7llama.cpp · Q2_0
- 18157.0tok/s65.7llama.cpp · Q2_0
- 19155.7tok/s66.0llama.cpp · Q2_0
- 20154.3tok/s57.6llama.cpp · Q2_0
- 21144.2tok/s46.8llama.cpp · Q2_0
- 22126.3tok/s20.2llama.cpp · Q2_0
Reproduce this run
toolllama.cpp manual
modelprism-ml/Ternary-Bonsai-2-9B-gguf:PQ2_0 (Q2_0)
context262,144
thinkingon
think limitnone
samplingtemperature 0.699999988079071 · top_p 0.949999988079071 · top_k 20 · min_p 0 · presence_penalty 0 · repeat_penalty 1
clientv0.4.96+102
Set those in llama.cpp, then:
llm-benchmark benchmark --model "prism-ml/Ternary-Bonsai-2-9B-gguf:PQ2_0" --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.