| { |
| "layerIndex": 0, |
| "artifactPath": "layers/layer_000.elasticbit", |
| "artifactBytes": 295173165, |
| "artifactSha256": "d23b1abbb08db2fd141e2d5c62b83efb19ee9987c61d7384329d8712e48013ca", |
| "artifactType": "elasticbit-block-root-v1", |
| "sourceLayerBytesFP16": 681594880, |
| "compressionSeconds": 5.484041452407837, |
| "threshold": 0.01, |
| "device": "cuda:0", |
| "elasticBitCudaDeviceIndex": 0, |
| "matrixSummary": { |
| "count": 4, |
| "histogram": { |
| "8": 3, |
| "6": 1 |
| }, |
| "effectiveBits": 7.769230769230769, |
| "originalBytes": 681574400, |
| "storageBytes": 331171840, |
| "storageReduction": 0.5141075721153846, |
| "maxMatrixError": 0.008025719784200191 |
| }, |
| "isolatedCalibrationValidation": { |
| "relativeL2": 0.030140706344435216, |
| "maxAbsDiff": 0.0458984375 |
| }, |
| "independentValidation": { |
| "relativeL2": 0.029015822065195637, |
| "maxAbsDiff": 0.0400390625 |
| }, |
| "artifactRoundTripValidation": { |
| "relativeL2": 0.0, |
| "maxAbsDiff": 0.0 |
| }, |
| "artifactRoundTripLowMem": { |
| "placeholderCount": 4, |
| "avoidedDenseBytesApproxFP16": 681574400, |
| "preloadedStateSlots": 2, |
| "movedNonPersistentBuffers": [], |
| "postloadExactStateSlots": 2, |
| "checkedStateDtypes": 2, |
| "dtypeMismatchCount": 0, |
| "stateValueMismatchCount": 0 |
| }, |
| "artifactRoundTripDevice": "cuda:0", |
| "matrices": [ |
| { |
| "name": "block.self_attn.o_proj", |
| "storageBits": 8, |
| "executionWidth": 8, |
| "error": 0.007830249145627022, |
| "storageBytes": 26234880, |
| "executionBytes": 26234880, |
| "originalBytes": 52428800, |
| "memoryReduction": 0.499609375 |
| }, |
| { |
| "name": "block.self_attn.qkv_proj", |
| "storageBits": 6, |
| "executionWidth": 8, |
| "error": 0.006366916466504335, |
| "storageBytes": 29521920, |
| "executionBytes": 39352320, |
| "originalBytes": 78643200, |
| "memoryReduction": 0.624609375 |
| }, |
| { |
| "name": "block.mlp.gate_up_proj", |
| "storageBits": 8, |
| "executionWidth": 8, |
| "error": 0.0062302532605826855, |
| "storageBytes": 183644160, |
| "executionBytes": 183644160, |
| "originalBytes": 367001600, |
| "memoryReduction": 0.499609375 |
| }, |
| { |
| "name": "block.mlp.down_proj", |
| "storageBits": 8, |
| "executionWidth": 8, |
| "error": 0.008025719784200191, |
| "storageBytes": 91770880, |
| "executionBytes": 91770880, |
| "originalBytes": 183500800, |
| "memoryReduction": 0.4998883928571428 |
| } |
| ] |
| } |
