commonllama benchmark · run
Run 0da4f7791ada
One run of Qwen3 4B on one machine class. This page is the durable record a submission receipt points to.
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.
Identity
What this record is.
| Run id | 0da4f7791ada25fa940445fbd50d59ed |
|---|---|
| Harness run id | run-73s99zqb48 |
| Content digest | bb0c650fdaa9f2994848fe8ff96477e544b11cdd9e6f2d8d76bf0277c3ccadd9 canonical, from the committed envelope |
| Raw byte digest | b5c5b2279cc5d86d50c8feec8a5fbbf6bda4dd31645463a82aec1dae29588f76 sha256 over the file bytes |
| Provenance | migration |
| Report status | complete |
Setting
Where the measurement ran.
| Hardware class | goldspier-rtx5090 |
|---|---|
| System RAM | unavailable bucket, taxonomy v1 |
| Backend | vulkan |
| Storage medium | ssd |
| Model | Qwen3 4B 3e4cb1417446 |
Qualified to at least 131,072 tokens of context, an observed lower bound, not a maximum.
Ladder
Every rung, every repetition.
Policy mode backend_default, ladder taxonomy v1. A cell that carries no time carries its typed reason instead.
| Context | Rep | Outcome | Actual tokens | Cook | Reload | Reason |
|---|---|---|---|---|---|---|
| 4,096 | 1 | ok | 4,096 | 633.5 ms | 9.99 s | |
| 8,192 | 1 | ok | 8,192 | 1.08 s | 11.96 s | |
| 16,384 | 1 | ok | 16,384 | 2.31 s | 16.78 s | |
| 32,768 | 1 | ok | 32,768 | 5.89 s | 26.05 s | |
| 65,536 | 1 | ok | 65,536 | 16.81 s | 45.99 s | |
| 131,072 | 1 | ok | 131,072 | 53.88 s | 82.48 s |
Saved context
Byte facts as recorded.
| Metric | Context | KV types | Availability | Bytes | Bytes per token | Source |
|---|---|---|---|---|---|---|
artifact-density | 4,096 | not recorded | available | unavailable | goldspier-f-checkpoint | |
artifact-density | 4,096 | not recorded | available | goldspier-f-resident | ||
artifact-density | 8,192 | not recorded | available | unavailable | goldspier-f-checkpoint | |
artifact-density | 8,192 | not recorded | available | goldspier-f-resident | ||
artifact-density | 16,384 | not recorded | available | unavailable | goldspier-f-checkpoint | |
artifact-density | 16,384 | not recorded | available | goldspier-f-resident | ||
artifact-density | 32,768 | not recorded | available | unavailable | goldspier-f-checkpoint | |
artifact-density | 32,768 | not recorded | available | goldspier-f-resident | ||
artifact-density | 65,536 | not recorded | available | unavailable | goldspier-f-checkpoint | |
artifact-density | 65,536 | not recorded | available | goldspier-f-resident | ||
artifact-density | 131,072 | not recorded | available | unavailable | goldspier-f-checkpoint | |
artifact-density | 131,072 | not recorded | available | goldspier-f-resident |
Runtime binding
The configuration the run bound to.
| Requested digest | not recorded |
|---|---|
| Effective digest | unavailable no reason recorded. |
Digests are stored as the harness emitted them, over the recursively key sorted public tuple. The site never recomputes one.
Software identities
The software behind the numbers.
These identities accompany every result so a stale row is visibly stale rather than quietly comparable.
| Harness | 0.0.0 |
|---|---|
| Source commit | 50dbc6c5512c87c9fbd0592628db07527b486e5a |
| Engine | llama.cpp |
| Addon | bf1b8c52dd6bde456aac056911a196a8fe75bb7518432613f8e3016d1b9852dc |
| Backend build | vulkan |
| Upstream | 082b326fc76f6e9bbb835b3920a3022bfdb6691c |
| Patchset | f0b3859395701d3d2b258de4e89ccff407e7c62e983f0636e6097a7b363e599b |
| Schema | commonllama.website-benchmark.v1 |
| Measurement contract | not recorded |
commonllama benchmark records are dedicated to the public domain under CC0 1.0. Use them for anything.