{ "layerIndex": 24, "artifactPath": "layers/layer_024.elasticbit", "artifactBytes": 331542116, "artifactSha256": "a7654f48b4618d66ef63d96599d8275b683a6a440cf4c8c76612605d3db20db6", "artifactType": "elasticbit-block-root-v1", "sourceLayerBytesFP16": 681594880, "compressionSeconds": 7.2748823165893555, "threshold": 0.01, "device": "cuda:1", "elasticBitCudaDeviceIndex": 1, "matrixSummary": { "count": 4, "histogram": { "9": 2, "7": 1, "8": 1 }, "effectiveBits": 8.23076923076923, "originalBytes": 681574400, "storageBytes": 350832640, "storageReduction": 0.48526141826923075, "maxMatrixError": 0.009108430705964565 }, "isolatedCalibrationValidation": { "relativeL2": 0.00300429056699515, "maxAbsDiff": 0.130859375 }, "independentValidation": { "relativeL2": 0.003127463498711889, "maxAbsDiff": 0.109375 }, "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": 0.005218408536165953, "storageBytes": 29511680, "executionBytes": 52428800, "originalBytes": 52428800, "memoryReduction": 0.437109375 }, { "name": "block.self_attn.qkv_proj", "storageBits": 7, "executionWidth": 8, "error": 0.009108430705964565, "storageBytes": 34437120, "executionBytes": 39352320, "originalBytes": 78643200, "memoryReduction": 0.562109375 }, { "name": "block.mlp.gate_up_proj", "storageBits": 8, "executionWidth": 8, "error": 0.0067471349611878395, "storageBytes": 183644160, "executionBytes": 183644160, "originalBytes": 367001600, "memoryReduction": 0.499609375 }, { "name": "block.mlp.down_proj", "storageBits": 9, "executionWidth": 16, "error": 0.005440438166260719, "storageBytes": 103239680, "executionBytes": 183500800, "originalBytes": 183500800, "memoryReduction": 0.4373883928571428 } ] }