ElasticBit layers 020-023 (atomic batch, nextLayer=24)

.gitattributes

@@ -53,3 +53,7 @@ layers/layer_016.elasticbit filter=lfs diff=lfs merge=lfs -text
 layers/layer_017.elasticbit filter=lfs diff=lfs merge=lfs -text
 layers/layer_018.elasticbit filter=lfs diff=lfs merge=lfs -text
 layers/layer_019.elasticbit filter=lfs diff=lfs merge=lfs -text
+layers/layer_020.elasticbit filter=lfs diff=lfs merge=lfs -text
+layers/layer_021.elasticbit filter=lfs diff=lfs merge=lfs -text
+layers/layer_022.elasticbit filter=lfs diff=lfs merge=lfs -text
+layers/layer_023.elasticbit filter=lfs diff=lfs merge=lfs -text

layers/layer_020.elasticbit

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:90d7ce6129ff978e3340563ec39e547f78788a1dfdfc67f9112eeb48fe36722a
+size 331373548

layers/layer_021.elasticbit

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:014289e43275d58925e4d2a6a547484eaf933b509c6d594f4e5688309bfb7531
+size 331435085

layers/layer_022.elasticbit

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2da85acefc79aaeca898f58636e0fca1a53c6726ca685f68214c82b3b1d91eaf
+size 331445332

layers/layer_023.elasticbit

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37b6bc590922a6c98a412762a8d59ad017d0514a5beff6e741866f9366b034c4
+size 331457953

progress.json

@@ -988,19 +988,211 @@
         "stateValueMismatchCount": 0
       },
       "artifactRoundTripDevice": "cuda:0"
+    },
+    {
+      "layerIndex": 20,
+      "artifactPath": "layers/layer_020.elasticbit",
+      "artifactBytes": 331373548,
+      "artifactSha256": "90d7ce6129ff978e3340563ec39e547f78788a1dfdfc67f9112eeb48fe36722a",
+      "artifactType": "elasticbit-block-root-v1",
+      "sourceLayerBytesFP16": 681594880,
+      "compressionSeconds": 7.181199550628662,
+      "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": 0.008326969109475613
+      },
+      "isolatedCalibrationValidation": {
+        "relativeL2": 0.0026045544263233284,
+        "maxAbsDiff": 0.1806640625
+      },
+      "independentValidation": {
+        "relativeL2": 0.002702716273242325,
+        "maxAbsDiff": 0.125
+      },
+      "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"
+    },
+    {
+      "layerIndex": 21,
+      "artifactPath": "layers/layer_021.elasticbit",
+      "artifactBytes": 331435085,
+      "artifactSha256": "014289e43275d58925e4d2a6a547484eaf933b509c6d594f4e5688309bfb7531",
+      "artifactType": "elasticbit-block-root-v1",
+      "sourceLayerBytesFP16": 681594880,
+      "compressionSeconds": 7.080434322357178,
+      "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": 0.008856349624693394
+      },
+      "isolatedCalibrationValidation": {
+        "relativeL2": 0.0027250093387543133,
+        "maxAbsDiff": 0.5
+      },
+      "independentValidation": {
+        "relativeL2": 0.002858023505256166,
+        "maxAbsDiff": 0.1171875
+      },
+      "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"
+    },
+    {
+      "layerIndex": 22,
+      "artifactPath": "layers/layer_022.elasticbit",
+      "artifactBytes": 331445332,
+      "artifactSha256": "2da85acefc79aaeca898f58636e0fca1a53c6726ca685f68214c82b3b1d91eaf",
+      "artifactType": "elasticbit-block-root-v1",
+      "sourceLayerBytesFP16": 681594880,
+      "compressionSeconds": 7.054975748062134,
+      "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": 0.008715600706636906
+      },
+      "isolatedCalibrationValidation": {
+        "relativeL2": 0.0028793881972116594,
+        "maxAbsDiff": 0.14453125
+      },
+      "independentValidation": {
+        "relativeL2": 0.002979165383702904,
+        "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"
+    },
+    {
+      "layerIndex": 23,
+      "artifactPath": "layers/layer_023.elasticbit",
+      "artifactBytes": 331457953,
+      "artifactSha256": "37b6bc590922a6c98a412762a8d59ad017d0514a5beff6e741866f9366b034c4",
+      "artifactType": "elasticbit-block-root-v1",
+      "sourceLayerBytesFP16": 681594880,
+      "compressionSeconds": 6.9463255405426025,
+      "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": 0.008419889025390148
+      },
+      "isolatedCalibrationValidation": {
+        "relativeL2": 0.003062929332325967,
+        "maxAbsDiff": 0.5
+      },
+      "independentValidation": {
+        "relativeL2": 0.0030691627060804687,
+        "maxAbsDiff": 0.1953125
+      },
+      "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"
     }
   ],
-  "nextLayer": 20,
+  "nextLayer": 24,
   "reconstructionSmoke": null,
   "complete": false,
   "aggregate": {
-    "completedLayers": 20,
-    "linearOriginalBytes": 13631488000,
-    "linearElasticBitStorageBytes": 6864281600,
-    "linearStorageReduction": 0.49643930288461535,
-    "parameterWeightedEffectiveBits": 8.051923076923076,
-    "layerArtifactBytes": 6430117997,
+    "completedLayers": 24,
+    "linearOriginalBytes": 16357785600,
+    "linearElasticBitStorageBytes": 8267612160,
+    "linearStorageReduction": 0.4945763221153846,
+    "parameterWeightedEffectiveBits": 8.08173076923077,
+    "layerArtifactBytes": 7755829915,
     "globalComponentBytes": 2055219848,
-    "totalUploadedModelBytesApprox": 8485337845
+    "totalUploadedModelBytesApprox": 9811049763
   }
 }

reports/layer_020.json

@@ -0,0 +1,90 @@
+{
+  "layerIndex": 20,
+  "artifactPath": "layers/layer_020.elasticbit",
+  "artifactBytes": 331373548,
+  "artifactSha256": "90d7ce6129ff978e3340563ec39e547f78788a1dfdfc67f9112eeb48fe36722a",
+  "artifactType": "elasticbit-block-root-v1",
+  "sourceLayerBytesFP16": 681594880,
+  "compressionSeconds": 7.181199550628662,
+  "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": 0.008326969109475613
+  },
+  "isolatedCalibrationValidation": {
+    "relativeL2": 0.0026045544263233284,
+    "maxAbsDiff": 0.1806640625
+  },
+  "independentValidation": {
+    "relativeL2": 0.002702716273242325,
+    "maxAbsDiff": 0.125
+  },
+  "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": 0.006358081940561533,
+      "storageBytes": 29511680,
+      "executionBytes": 52428800,
+      "originalBytes": 52428800,
+      "memoryReduction": 0.437109375
+    },
+    {
+      "name": "block.self_attn.qkv_proj",
+      "storageBits": 7,
+      "executionWidth": 8,
+      "error": 0.008326969109475613,
+      "storageBytes": 34437120,
+      "executionBytes": 39352320,
+      "originalBytes": 78643200,
+      "memoryReduction": 0.562109375
+    },
+    {
+      "name": "block.mlp.gate_up_proj",
+      "storageBits": 8,
+      "executionWidth": 8,
+      "error": 0.006741302087903023,
+      "storageBytes": 183644160,
+      "executionBytes": 183644160,
+      "originalBytes": 367001600,
+      "memoryReduction": 0.499609375
+    },
+    {
+      "name": "block.mlp.down_proj",
+      "storageBits": 9,
+      "executionWidth": 16,
+      "error": 0.006001586560159922,
+      "storageBytes": 103239680,
+      "executionBytes": 183500800,
+      "originalBytes": 183500800,
+      "memoryReduction": 0.4373883928571428
+    }
+  ]
+}

reports/layer_021.json

@@ -0,0 +1,90 @@
+{
+  "layerIndex": 21,
+  "artifactPath": "layers/layer_021.elasticbit",
+  "artifactBytes": 331435085,
+  "artifactSha256": "014289e43275d58925e4d2a6a547484eaf933b509c6d594f4e5688309bfb7531",
+  "artifactType": "elasticbit-block-root-v1",
+  "sourceLayerBytesFP16": 681594880,
+  "compressionSeconds": 7.080434322357178,
+  "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": 0.008856349624693394
+  },
+  "isolatedCalibrationValidation": {
+    "relativeL2": 0.0027250093387543133,
+    "maxAbsDiff": 0.5
+  },
+  "independentValidation": {
+    "relativeL2": 0.002858023505256166,
+    "maxAbsDiff": 0.1171875
+  },
+  "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": 0.006340627558529377,
+      "storageBytes": 29511680,
+      "executionBytes": 52428800,
+      "originalBytes": 52428800,
+      "memoryReduction": 0.437109375
+    },
+    {
+      "name": "block.self_attn.qkv_proj",
+      "storageBits": 7,
+      "executionWidth": 8,
+      "error": 0.008856349624693394,
+      "storageBytes": 34437120,
+      "executionBytes": 39352320,
+      "originalBytes": 78643200,
+      "memoryReduction": 0.562109375
+    },
+    {
+      "name": "block.mlp.gate_up_proj",
+      "storageBits": 8,
+      "executionWidth": 8,
+      "error": 0.00669582886621356,
+      "storageBytes": 183644160,
+      "executionBytes": 183644160,
+      "originalBytes": 367001600,
+      "memoryReduction": 0.499609375
+    },
+    {
+      "name": "block.mlp.down_proj",
+      "storageBits": 9,
+      "executionWidth": 16,
+      "error": 0.005525866989046335,
+      "storageBytes": 103239680,
+      "executionBytes": 183500800,
+      "originalBytes": 183500800,
+      "memoryReduction": 0.4373883928571428
+    }
+  ]
+}

reports/layer_022.json

@@ -0,0 +1,90 @@
+{
+  "layerIndex": 22,
+  "artifactPath": "layers/layer_022.elasticbit",
+  "artifactBytes": 331445332,
+  "artifactSha256": "2da85acefc79aaeca898f58636e0fca1a53c6726ca685f68214c82b3b1d91eaf",
+  "artifactType": "elasticbit-block-root-v1",
+  "sourceLayerBytesFP16": 681594880,
+  "compressionSeconds": 7.054975748062134,
+  "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": 0.008715600706636906
+  },
+  "isolatedCalibrationValidation": {
+    "relativeL2": 0.0028793881972116594,
+    "maxAbsDiff": 0.14453125
+  },
+  "independentValidation": {
+    "relativeL2": 0.002979165383702904,
+    "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": 0.007206959184259176,
+      "storageBytes": 29511680,
+      "executionBytes": 52428800,
+      "originalBytes": 52428800,
+      "memoryReduction": 0.437109375
+    },
+    {
+      "name": "block.self_attn.qkv_proj",
+      "storageBits": 7,
+      "executionWidth": 8,
+      "error": 0.008715600706636906,
+      "storageBytes": 34437120,
+      "executionBytes": 39352320,
+      "originalBytes": 78643200,
+      "memoryReduction": 0.562109375
+    },
+    {
+      "name": "block.mlp.gate_up_proj",
+      "storageBits": 8,
+      "executionWidth": 8,
+      "error": 0.00658106105402112,
+      "storageBytes": 183644160,
+      "executionBytes": 183644160,
+      "originalBytes": 367001600,
+      "memoryReduction": 0.499609375
+    },
+    {
+      "name": "block.mlp.down_proj",
+      "storageBits": 9,
+      "executionWidth": 16,
+      "error": 0.00556604377925396,
+      "storageBytes": 103239680,
+      "executionBytes": 183500800,
+      "originalBytes": 183500800,
+      "memoryReduction": 0.4373883928571428
+    }
+  ]
+}

reports/layer_023.json

@@ -0,0 +1,90 @@
+{
+  "layerIndex": 23,
+  "artifactPath": "layers/layer_023.elasticbit",
+  "artifactBytes": 331457953,
+  "artifactSha256": "37b6bc590922a6c98a412762a8d59ad017d0514a5beff6e741866f9366b034c4",
+  "artifactType": "elasticbit-block-root-v1",
+  "sourceLayerBytesFP16": 681594880,
+  "compressionSeconds": 6.9463255405426025,
+  "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": 0.008419889025390148
+  },
+  "isolatedCalibrationValidation": {
+    "relativeL2": 0.003062929332325967,
+    "maxAbsDiff": 0.5
+  },
+  "independentValidation": {
+    "relativeL2": 0.0030691627060804687,
+    "maxAbsDiff": 0.1953125
+  },
+  "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": 0.005984076298773289,
+      "storageBytes": 29511680,
+      "executionBytes": 52428800,
+      "originalBytes": 52428800,
+      "memoryReduction": 0.437109375
+    },
+    {
+      "name": "block.self_attn.qkv_proj",
+      "storageBits": 7,
+      "executionWidth": 8,
+      "error": 0.008419889025390148,
+      "storageBytes": 34437120,
+      "executionBytes": 39352320,
+      "originalBytes": 78643200,
+      "memoryReduction": 0.562109375
+    },
+    {
+      "name": "block.mlp.gate_up_proj",
+      "storageBits": 8,
+      "executionWidth": 8,
+      "error": 0.006554744206368923,
+      "storageBytes": 183644160,
+      "executionBytes": 183644160,
+      "originalBytes": 367001600,
+      "memoryReduction": 0.499609375
+    },
+    {
+      "name": "block.mlp.down_proj",
+      "storageBits": 9,
+      "executionWidth": 16,
+      "error": 0.005664248019456863,
+      "storageBytes": 103239680,
+      "executionBytes": 183500800,
+      "originalBytes": 183500800,
+      "memoryReduction": 0.4373883928571428
+    }
+  ]
+}