commonllama benchmark · model
Granite 4.0 H Tiny
Every number on this page is an observed measurement from a benchmark run, shown with the run that produced it.
Pre-release: these numbers come from commonllama harness 0.0.0, a build not yet released. They are real measurements, not estimates; released-harness numbers will replace them.
Observed facts
What the harness read from the file.
| Model file digest | 064bea0136420b38d0b65697fa5e772e28b112eee1757aacc7f64eba6bf37810 |
|---|---|
| Architecture | granitehybrid |
| Quantization | q4_k_m |
| Max context | 1,048,576 tokens |
| Model file size | 3.94 GiB (4,230,975,936 bytes) |
| Fact provenance | unknown |
Context capacity
How much context this model has carried.
Qualified to at least 131,072 tokens of context, an observed lower bound, not a maximum.
Observed on darwin-arm64-apple-accelerator metal. A frontier is a property of one run on one machine class, never a promise about another.
Saved context
What a saved context costs on disk.
Each curve keeps to one named key: this model, one backend. Unlike measurements are never folded together.
The KV cache is the AI’s memory of the conversation so far. Saving it to disk lets the AI return to the conversation and load that memory back in, rather than read everything from the start again.
| Backend | Context | Bytes per token | Saved bytes |
|---|---|---|---|
| cpu | 4,096 | unavailable | unavailable |
| cpu | 8,192 | unavailable | unavailable |
| cpu | 16,384 | unavailable | unavailable |
| cpu | 32,768 | unavailable | unavailable |
| cpu | 65,536 | unavailable | unavailable |
| cpu | 131,072 | unavailable | unavailable |
| cpu | 4,096 | 75,936,752 / 4,096 (~18,539.2 bytes per token) | 72.4 MiB (75,936,752 bytes) |
| cpu | 8,192 | 93,811,696 / 8,192 (~11,451.6 bytes per token) | 89.5 MiB (93,811,696 bytes) |
| cpu | 16,384 | 129,561,584 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,561,584 bytes) |
| cpu | 32,768 | 201,061,360 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,360 bytes) |
| cpu | 65,536 | 344,060,912 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,060,912 bytes) |
| cpu | 131,072 | 630,060,016 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,016 bytes) |
| cuda | 4,096 | unavailable | unavailable |
| cuda | 4,096 | unavailable | unavailable |
| cuda | 4,096 | unavailable | unavailable |
| cuda | 4,096 | unavailable | unavailable |
| cuda | 8,192 | unavailable | unavailable |
| cuda | 8,192 | unavailable | unavailable |
| cuda | 8,192 | unavailable | unavailable |
| cuda | 8,192 | unavailable | unavailable |
| cuda | 16,384 | unavailable | unavailable |
| cuda | 16,384 | unavailable | unavailable |
| cuda | 16,384 | unavailable | unavailable |
| cuda | 16,384 | unavailable | unavailable |
| cuda | 32,768 | unavailable | unavailable |
| cuda | 32,768 | unavailable | unavailable |
| cuda | 32,768 | unavailable | unavailable |
| cuda | 32,768 | unavailable | unavailable |
| cuda | 65,536 | unavailable | unavailable |
| cuda | 65,536 | unavailable | unavailable |
| cuda | 65,536 | unavailable | unavailable |
| cuda | 65,536 | unavailable | unavailable |
| cuda | 131,072 | unavailable | unavailable |
| cuda | 131,072 | unavailable | unavailable |
| cuda | 131,072 | unavailable | unavailable |
| cuda | 131,072 | unavailable | unavailable |
| cuda | 4,096 | 75,936,752 / 4,096 (~18,539.2 bytes per token) | 72.4 MiB (75,936,752 bytes) |
| cuda | 4,096 | unavailable | unavailable |
| cuda | 4,096 | 75,936,752 / 4,096 (~18,539.2 bytes per token) | 72.4 MiB (75,936,752 bytes) |
| cuda | 4,096 | unavailable | unavailable |
| cuda | 8,192 | 93,811,696 / 8,192 (~11,451.6 bytes per token) | 89.5 MiB (93,811,696 bytes) |
| cuda | 8,192 | unavailable | unavailable |
| cuda | 8,192 | 93,811,696 / 8,192 (~11,451.6 bytes per token) | 89.5 MiB (93,811,696 bytes) |
| cuda | 8,192 | unavailable | unavailable |
| cuda | 16,384 | 129,561,584 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,561,584 bytes) |
| cuda | 16,384 | unavailable | unavailable |
| cuda | 16,384 | 129,561,584 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,561,584 bytes) |
| cuda | 16,384 | unavailable | unavailable |
| cuda | 32,768 | 201,061,360 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,360 bytes) |
| cuda | 32,768 | unavailable | unavailable |
| cuda | 32,768 | 201,061,360 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,360 bytes) |
| cuda | 32,768 | unavailable | unavailable |
| cuda | 65,536 | 344,060,912 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,060,912 bytes) |
| cuda | 65,536 | unavailable | unavailable |
| cuda | 65,536 | 344,060,912 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,060,912 bytes) |
| cuda | 65,536 | unavailable | unavailable |
| cuda | 131,072 | 630,060,016 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,016 bytes) |
| cuda | 131,072 | unavailable | unavailable |
| cuda | 131,072 | 630,060,016 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,016 bytes) |
| cuda | 131,072 | unavailable | unavailable |
| metal | 4,096 | unavailable | unavailable |
| metal | 4,096 | unavailable | unavailable |
| metal | 4,096 | unavailable | unavailable |
| metal | 8,192 | unavailable | unavailable |
| metal | 8,192 | unavailable | unavailable |
| metal | 8,192 | unavailable | unavailable |
| metal | 16,384 | unavailable | unavailable |
| metal | 16,384 | unavailable | unavailable |
| metal | 16,384 | unavailable | unavailable |
| metal | 32,768 | unavailable | unavailable |
| metal | 32,768 | unavailable | unavailable |
| metal | 32,768 | unavailable | unavailable |
| metal | 65,536 | unavailable | unavailable |
| metal | 65,536 | unavailable | unavailable |
| metal | 65,536 | unavailable | unavailable |
| metal | 131,072 | unavailable | unavailable |
| metal | 131,072 | unavailable | unavailable |
| metal | 131,072 | unavailable | unavailable |
| metal | 4,096 | 75,937,315 / 4,096 (~18,539.4 bytes per token) | 72.4 MiB (75,937,315 bytes) |
| metal | 4,096 | 75,937,315 / 4,096 (~18,539.4 bytes per token) | 72.4 MiB (75,937,315 bytes) |
| metal | 4,096 | 75,937,315 / 4,096 (~18,539.4 bytes per token) | 72.4 MiB (75,937,315 bytes) |
| metal | 8,192 | 93,812,259 / 8,192 (~11,451.7 bytes per token) | 89.5 MiB (93,812,259 bytes) |
| metal | 8,192 | 93,812,259 / 8,192 (~11,451.7 bytes per token) | 89.5 MiB (93,812,259 bytes) |
| metal | 8,192 | 93,812,259 / 8,192 (~11,451.7 bytes per token) | 89.5 MiB (93,812,259 bytes) |
| metal | 16,384 | 129,562,147 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,562,147 bytes) |
| metal | 16,384 | 129,562,147 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,562,147 bytes) |
| metal | 16,384 | 129,562,147 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,562,147 bytes) |
| metal | 32,768 | 201,061,923 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,923 bytes) |
| metal | 32,768 | 201,061,923 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,923 bytes) |
| metal | 32,768 | 201,061,923 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,923 bytes) |
| metal | 65,536 | 344,061,475 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,061,475 bytes) |
| metal | 65,536 | 344,061,475 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,061,475 bytes) |
| metal | 65,536 | 344,061,475 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,061,475 bytes) |
| metal | 131,072 | 630,060,579 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,579 bytes) |
| metal | 131,072 | 630,060,579 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,579 bytes) |
| metal | 131,072 | 630,060,579 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,579 bytes) |
| vulkan | 4,096 | unavailable | unavailable |
| vulkan | 4,096 | unavailable | unavailable |
| vulkan | 4,096 | unavailable | unavailable |
| vulkan | 4,096 | unavailable | unavailable |
| vulkan | 8,192 | unavailable | unavailable |
| vulkan | 8,192 | unavailable | unavailable |
| vulkan | 8,192 | unavailable | unavailable |
| vulkan | 8,192 | unavailable | unavailable |
| vulkan | 16,384 | unavailable | unavailable |
| vulkan | 16,384 | unavailable | unavailable |
| vulkan | 16,384 | unavailable | unavailable |
| vulkan | 16,384 | unavailable | unavailable |
| vulkan | 32,768 | unavailable | unavailable |
| vulkan | 32,768 | unavailable | unavailable |
| vulkan | 32,768 | unavailable | unavailable |
| vulkan | 32,768 | unavailable | unavailable |
| vulkan | 65,536 | unavailable | unavailable |
| vulkan | 65,536 | unavailable | unavailable |
| vulkan | 65,536 | unavailable | unavailable |
| vulkan | 65,536 | unavailable | unavailable |
| vulkan | 131,072 | unavailable | unavailable |
| vulkan | 131,072 | unavailable | unavailable |
| vulkan | 131,072 | unavailable | unavailable |
| vulkan | 131,072 | unavailable | unavailable |
| vulkan | 4,096 | 75,937,315 / 4,096 (~18,539.4 bytes per token) | 72.4 MiB (75,937,315 bytes) |
| vulkan | 4,096 | unavailable | unavailable |
| vulkan | 4,096 | 75,936,752 / 4,096 (~18,539.2 bytes per token) | 72.4 MiB (75,936,752 bytes) |
| vulkan | 4,096 | unavailable | unavailable |
| vulkan | 8,192 | 93,812,259 / 8,192 (~11,451.7 bytes per token) | 89.5 MiB (93,812,259 bytes) |
| vulkan | 8,192 | unavailable | unavailable |
| vulkan | 8,192 | 93,811,696 / 8,192 (~11,451.6 bytes per token) | 89.5 MiB (93,811,696 bytes) |
| vulkan | 8,192 | unavailable | unavailable |
| vulkan | 16,384 | 129,562,147 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,562,147 bytes) |
| vulkan | 16,384 | unavailable | unavailable |
| vulkan | 16,384 | 129,561,584 / 16,384 (~7,907.8 bytes per token) | 123.6 MiB (129,561,584 bytes) |
| vulkan | 16,384 | unavailable | unavailable |
| vulkan | 32,768 | 201,061,923 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,923 bytes) |
| vulkan | 32,768 | unavailable | unavailable |
| vulkan | 32,768 | 201,061,360 / 32,768 (~6,135.9 bytes per token) | 191.7 MiB (201,061,360 bytes) |
| vulkan | 32,768 | unavailable | unavailable |
| vulkan | 65,536 | 344,061,475 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,061,475 bytes) |
| vulkan | 65,536 | unavailable | unavailable |
| vulkan | 65,536 | 344,060,912 / 65,536 (~5,250.0 bytes per token) | 328.1 MiB (344,060,912 bytes) |
| vulkan | 65,536 | unavailable | unavailable |
| vulkan | 131,072 | 630,060,579 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,579 bytes) |
| vulkan | 131,072 | unavailable | unavailable |
| vulkan | 131,072 | 630,060,016 / 131,072 (~4,807.0 bytes per token) | 600.9 MiB (630,060,016 bytes) |
| vulkan | 131,072 | unavailable | unavailable |
Provenance
Every run behind the numbers.
| Run | Provenance | Status | Hardware class | RAM bucket | Backend | Storage | Frontier |
|---|---|---|---|---|---|---|---|
967cf195f13c | community | complete | darwin-arm64-apple-accelerator | 16 to 31 GiB | metal | other | 131,072+ |
351b00ef1aa6 | community | complete | windows-x64-nvidia-accelerator | 32 to 63 GiB | vulkan | other | 131,072+ |
53194a4bd2c1 | community | complete | darwin-arm64-apple-accelerator | 16 to 31 GiB | metal | other | 131,072+ |
71290f56840b | community | complete | darwin-arm64-apple-accelerator | 16 to 31 GiB | metal | other | 131,072+ |
3850ccf6d4e1 | migration | complete | goldspier-x64-cuda | unavailable | vulkan | resident-memory | 131,072+ |
58dfd141f342 | migration | complete | goldspier-x64-vulkan | unavailable | vulkan | resident-memory | 131,072+ |
674c59680555 | migration | complete | goldspier-x64-cuda | unavailable | cuda | resident-memory | 131,072+ |
6fbee9db97a8 | migration | complete | goldspier-x64-cuda | unavailable | cuda | resident-memory | 131,072+ |
a7053fabf3de | migration | complete | goldspier-x64-cuda | unavailable | cuda | resident-memory | 131,072+ |
ae84204c6426 | migration | complete | goldspier-x64-vulkan | unavailable | vulkan | resident-memory | 131,072+ |
af402ed27975 | migration | complete | goldspier-x64-cpu | unavailable | cpu · provisional at measurement | resident-memory | 131,072+ |
fd000c6827df | migration | complete | goldspier-x64-cuda | unavailable | cuda | resident-memory | 131,072+ |
A partial run is content, not a failure: its rungs carry typed reasons on the run page. Partial and complete runs are never mixed inside one aggregate on this page.
commonllama benchmark records are dedicated to the public domain under CC0 1.0. Use them for anything.