| { |
| "layerIndex": 1, |
| "artifactPath": "layers/layer_001.elasticbit", |
| "artifactBytes": 301655521, |
| "artifactSha256": "8480615b51df4cc4f7d052dadc8261af336912ed6fa3270d2a4b5246dbbba48d", |
| "artifactType": "elasticbit-block-root-v1", |
| "sourceLayerBytesFP16": 681594880, |
| "compressionSeconds": 5.89387321472168, |
| "threshold": 0.01, |
| "device": "cuda:0", |
| "elasticBitCudaDeviceIndex": 0, |
| "matrixSummary": { |
| "count": 4, |
| "histogram": { |
| "9": 1, |
| "6": 1, |
| "8": 1, |
| "7": 1 |
| }, |
| "effectiveBits": 7.576923076923077, |
| "originalBytes": 681574400, |
| "storageBytes": 322979840, |
| "storageReduction": 0.5261268028846153, |
| "maxMatrixError": 0.009119924157857895 |
| }, |
| "isolatedCalibrationValidation": { |
| "relativeL2": 0.008968063871791896, |
| "maxAbsDiff": 2.5234375 |
| }, |
| "independentValidation": { |
| "relativeL2": 0.00940780515518235, |
| "maxAbsDiff": 2.421875 |
| }, |
| "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": 9, |
| "executionWidth": 16, |
| "error": 0.005272475071251392, |
| "storageBytes": 29511680, |
| "executionBytes": 52428800, |
| "originalBytes": 52428800, |
| "memoryReduction": 0.437109375 |
| }, |
| { |
| "name": "block.self_attn.qkv_proj", |
| "storageBits": 6, |
| "executionWidth": 8, |
| "error": 0.009119924157857895, |
| "storageBytes": 29521920, |
| "executionBytes": 39352320, |
| "originalBytes": 78643200, |
| "memoryReduction": 0.624609375 |
| }, |
| { |
| "name": "block.mlp.gate_up_proj", |
| "storageBits": 8, |
| "executionWidth": 8, |
| "error": 0.006299629341810942, |
| "storageBytes": 183644160, |
| "executionBytes": 183644160, |
| "originalBytes": 367001600, |
| "memoryReduction": 0.499609375 |
| }, |
| { |
| "name": "block.mlp.down_proj", |
| "storageBits": 7, |
| "executionWidth": 8, |
| "error": 0.00860297679901123, |
| "storageBytes": 80302080, |
| "executionBytes": 91770880, |
| "originalBytes": 183500800, |
| "memoryReduction": 0.5623883928571429 |
| } |
| ] |
| } |
