commonllama benchmark · model

Qwen3 4B

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 digest3e4cb14174460404e7a233e531675303b2fbf7749c02f91864fe311ab6344e4f
Architecturenot observed
Quantizationnot observed
Max contextnot observed
Model file size2.33 GiB (2,497,280,480 bytes)
Fact provenanceunknown

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 nucbox-evo-x2 vulkan. 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.

Bytes per cached token, complete runs only
BackendContextBytes per tokenSaved bytes
cuda4,096unavailableunavailable
cuda8,192unavailableunavailable
cuda16,384unavailableunavailable
cuda32,768unavailableunavailable
cuda65,536unavailableunavailable
cuda131,072unavailableunavailable
cuda4,096unavailableunavailable
cuda8,192unavailableunavailable
cuda16,384unavailableunavailable
cuda32,768unavailableunavailable
cuda65,536unavailableunavailable
cuda131,072unavailableunavailable
cuda4,096unavailableunavailable
cuda4,096unavailableunavailable
cuda8,192unavailableunavailable
cuda8,192unavailableunavailable
cuda16,384unavailableunavailable
cuda16,384unavailableunavailable
cuda32,768unavailableunavailable
cuda32,768unavailableunavailable
cuda65,536unavailableunavailable
cuda65,536unavailableunavailable
cuda131,072unavailableunavailable
cuda131,072unavailableunavailable
rocm4,096unavailableunavailable
rocm4,096unavailableunavailable
rocm8,192unavailableunavailable
rocm8,192unavailableunavailable
rocm16,384unavailableunavailable
rocm16,384unavailableunavailable
rocm32,768unavailableunavailable
rocm32,768unavailableunavailable
rocm65,536unavailableunavailable
rocm65,536unavailableunavailable
rocm131,072unavailableunavailable
rocm131,072unavailableunavailable
rocm4,096320,835,948 / 4,096 (~78,329.1 bytes per token)306.0 MiB (320,835,948 bytes)
rocm4,096320,835,948 / 4,096 (~78,329.1 bytes per token)306.0 MiB (320,835,948 bytes)
rocm8,192641,749,356 / 8,192 (~78,338.5 bytes per token)612.0 MiB (641,749,356 bytes)
rocm8,192641,749,356 / 8,192 (~78,338.5 bytes per token)612.0 MiB (641,749,356 bytes)
rocm16,3841,283,576,172 / 16,384 (~78,343.3 bytes per token)1.20 GiB (1,283,576,172 bytes)
rocm16,3841,283,576,172 / 16,384 (~78,343.3 bytes per token)1.20 GiB (1,283,576,172 bytes)
rocm32,7682,567,229,804 / 32,768 (~78,345.6 bytes per token)2.39 GiB (2,567,229,804 bytes)
rocm32,7682,567,229,804 / 32,768 (~78,345.6 bytes per token)2.39 GiB (2,567,229,804 bytes)
rocm65,5365,134,537,068 / 65,536 (~78,346.8 bytes per token)4.78 GiB (5,134,537,068 bytes)
rocm65,5365,134,537,068 / 65,536 (~78,346.8 bytes per token)4.78 GiB (5,134,537,068 bytes)
rocm131,07210,269,151,596 / 131,072 (~78,347.4 bytes per token)9.56 GiB (10,269,151,596 bytes)
rocm131,07210,269,151,596 / 131,072 (~78,347.4 bytes per token)9.56 GiB (10,269,151,596 bytes)
vulkan4,096unavailableunavailable
vulkan8,192unavailableunavailable
vulkan16,384unavailableunavailable
vulkan32,768unavailableunavailable
vulkan65,536unavailableunavailable
vulkan131,072unavailableunavailable
vulkan4,096unavailableunavailable
vulkan8,192unavailableunavailable
vulkan16,384unavailableunavailable
vulkan32,768unavailableunavailable
vulkan65,536unavailableunavailable
vulkan131,072unavailableunavailable
vulkan4,096unavailableunavailable
vulkan8,192unavailableunavailable
vulkan16,384unavailableunavailable
vulkan32,768unavailableunavailable
vulkan65,536unavailableunavailable
vulkan131,072unavailableunavailable
vulkan4,096unavailableunavailable
vulkan4,096unavailableunavailable
vulkan8,192unavailableunavailable
vulkan8,192unavailableunavailable
vulkan16,384unavailableunavailable
vulkan16,384unavailableunavailable
vulkan32,768unavailableunavailable
vulkan32,768unavailableunavailable
vulkan65,536unavailableunavailable
vulkan65,536unavailableunavailable
vulkan131,072unavailableunavailable
vulkan131,072unavailableunavailable
vulkan4,096unavailableunavailable
vulkan4,096unavailableunavailable
vulkan4,096unavailableunavailable
vulkan4,096unavailableunavailable
vulkan8,192unavailableunavailable
vulkan8,192unavailableunavailable
vulkan8,192unavailableunavailable
vulkan8,192unavailableunavailable
vulkan16,384unavailableunavailable
vulkan16,384unavailableunavailable
vulkan16,384unavailableunavailable
vulkan16,384unavailableunavailable
vulkan32,768unavailableunavailable
vulkan32,768unavailableunavailable
vulkan32,768unavailableunavailable
vulkan32,768unavailableunavailable
vulkan65,536unavailableunavailable
vulkan65,536unavailableunavailable
vulkan65,536unavailableunavailable
vulkan65,536unavailableunavailable
vulkan131,072unavailableunavailable
vulkan131,072unavailableunavailable
vulkan131,072unavailableunavailable
vulkan131,072unavailableunavailable
vulkan4,096320,835,948 / 4,096 (~78,329.1 bytes per token)306.0 MiB (320,835,948 bytes)
vulkan4,096320,835,948 / 4,096 (~78,329.1 bytes per token)306.0 MiB (320,835,948 bytes)
vulkan4,096320,835,948 / 4,096 (~78,329.1 bytes per token)306.0 MiB (320,835,948 bytes)
vulkan4,096320,835,948 / 4,096 (~78,329.1 bytes per token)306.0 MiB (320,835,948 bytes)
vulkan8,192641,749,356 / 8,192 (~78,338.5 bytes per token)612.0 MiB (641,749,356 bytes)
vulkan8,192641,749,356 / 8,192 (~78,338.5 bytes per token)612.0 MiB (641,749,356 bytes)
vulkan8,192641,749,356 / 8,192 (~78,338.5 bytes per token)612.0 MiB (641,749,356 bytes)
vulkan8,192641,749,356 / 8,192 (~78,338.5 bytes per token)612.0 MiB (641,749,356 bytes)
vulkan16,3841,283,576,172 / 16,384 (~78,343.3 bytes per token)1.20 GiB (1,283,576,172 bytes)
vulkan16,3841,283,576,172 / 16,384 (~78,343.3 bytes per token)1.20 GiB (1,283,576,172 bytes)
vulkan16,3841,283,576,172 / 16,384 (~78,343.3 bytes per token)1.20 GiB (1,283,576,172 bytes)
vulkan16,3841,283,576,172 / 16,384 (~78,343.3 bytes per token)1.20 GiB (1,283,576,172 bytes)
vulkan32,7682,567,229,804 / 32,768 (~78,345.6 bytes per token)2.39 GiB (2,567,229,804 bytes)
vulkan32,7682,567,229,804 / 32,768 (~78,345.6 bytes per token)2.39 GiB (2,567,229,804 bytes)
vulkan32,7682,567,229,804 / 32,768 (~78,345.6 bytes per token)2.39 GiB (2,567,229,804 bytes)
vulkan32,7682,567,229,804 / 32,768 (~78,345.6 bytes per token)2.39 GiB (2,567,229,804 bytes)
vulkan65,5365,134,537,068 / 65,536 (~78,346.8 bytes per token)4.78 GiB (5,134,537,068 bytes)
vulkan65,5365,134,537,068 / 65,536 (~78,346.8 bytes per token)4.78 GiB (5,134,537,068 bytes)
vulkan65,5365,134,537,068 / 65,536 (~78,346.8 bytes per token)4.78 GiB (5,134,537,068 bytes)
vulkan65,5365,134,537,068 / 65,536 (~78,346.8 bytes per token)4.78 GiB (5,134,537,068 bytes)
vulkan131,07210,269,151,596 / 131,072 (~78,347.4 bytes per token)9.56 GiB (10,269,151,596 bytes)
vulkan131,07210,269,151,596 / 131,072 (~78,347.4 bytes per token)9.56 GiB (10,269,151,596 bytes)
vulkan131,07210,269,151,596 / 131,072 (~78,347.4 bytes per token)9.56 GiB (10,269,151,596 bytes)
vulkan131,07210,269,151,596 / 131,072 (~78,347.4 bytes per token)9.56 GiB (10,269,151,596 bytes)
vulkan4,096320,835,948 / 4,096 (~78,329.1 bytes per token)306.0 MiB (320,835,948 bytes)
vulkan8,192641,749,356 / 8,192 (~78,338.5 bytes per token)612.0 MiB (641,749,356 bytes)
vulkan16,3841,283,576,172 / 16,384 (~78,343.3 bytes per token)1.20 GiB (1,283,576,172 bytes)
vulkan32,7682,567,229,804 / 32,768 (~78,345.6 bytes per token)2.39 GiB (2,567,229,804 bytes)
vulkan65,5365,134,537,068 / 65,536 (~78,346.8 bytes per token)4.78 GiB (5,134,537,068 bytes)
vulkan131,07210,269,151,596 / 131,072 (~78,347.4 bytes per token)9.56 GiB (10,269,151,596 bytes)

Provenance

Every run behind the numbers.

Published runs
RunProvenanceStatusHardware classRAM bucketBackendStorageFrontier
012913e96242migrationcompletenucbox-evo-x2unavailablevulkanssd131,072+
0da4f7791adamigrationcompletegoldspier-rtx5090unavailablevulkanssd131,072+
36acee4641f7migrationcompletegoldspier-rtx5090unavailablecudassd131,072+
633fc4ef3155migrationcompletenucbox-evo-x2unavailablevulkanssd131,072+
83b0b2d70ccemigrationcompletegoldspier-rtx5090unavailablecudassd131,072+
a1a6c154b678migrationcompletenucbox-evo-x2unavailablevulkanssd131,072+
a54d5b9d273emigrationcompletenucbox-evo-x2unavailablerocm · provisional at measurementssd131,072+
b1541de848f6migrationcompletenucbox-evo-x2unavailablevulkanssd131,072+
b1c6d93c64c9migrationcompletenucbox-evo-x2unavailablerocm · provisional at measurementssd131,072+
f5b04a451b8bmigrationcompletegoldspier-rtx5090unavailablevulkanssd131,072+
f97a539d6ad1migrationcompletecommonllama-linux-x64-vulkan-6fecf59f2068ba93unavailablevulkanresident-memory131,072+

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.