GLiNER-Relex runtime artifact: FP16 key-only attention mask (correct on CPU/GPU as well as ANE)

GLiNERRelexBase.mlmodelc/analytics/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4fca3ff909bfe582309824ca6fb3aeb90bce25d3cd43badaa14c5ec6ae3ecdce
+oid sha256:11e9f41bbaafca94ca230b32c4654d562ecbdfd5f7f3f6c79493719145a51463
 size 243

GLiNERRelexBase.mlmodelc/coremldata.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c16ad119bbb2c82c57dd9706e72b212319633c210c84399c3c7916e7480b808
+oid sha256:80cea4ea64ee70c20549cc7a21cd053019c118154c686df8bd1309eb8e4e6d3a
 size 482

GLiNERRelexBase.mlmodelc/metadata.json

@@ -29,31 +29,30 @@
     ],
     "specificationVersion" : 9,
     "mlProgramOperationTypeHistogram" : {
-      "Ios18.expandDims" : 11,
+      "Ios18.expandDims" : 9,
       "Ios18.softmax" : 12,
-      "Ios18.mul" : 44,
+      "Ios18.mul" : 43,
       "Ios18.matmul" : 52,
       "Ios18.logicalNot" : 1,
       "Ios18.equal" : 5,
       "Ios18.gatherAlongAxis" : 28,
       "Select" : 12,
-      "Tile" : 6,
+      "Tile" : 7,
       "Ios18.gather" : 3,
       "Ios18.add" : 48,
       "Ios18.layerNorm" : 25,
-      "Ios18.reshape" : 122,
+      "Ios18.reshape" : 123,
       "Ios18.linear" : 86,
       "Ios18.gelu" : 12,
       "Ios18.lstm" : 2,
       "Ios18.sub" : 2,
       "Ios18.transpose" : 78,
-      "Ios18.cast" : 19,
+      "Ios18.cast" : 17,
       "Ios18.clip" : 1,
       "Ios18.concat" : 4,
       "Ios18.relu" : 7,
-      "Ios18.squeeze" : 1,
-      "Ios16.cumsum" : 2,
-      "Ios18.sliceByIndex" : 6
+      "Ios18.sliceByIndex" : 6,
+      "Ios16.cumsum" : 2
     },
     "computePrecision" : "Mixed (Float16, Float32, Int16, Int32, UInt16)",
     "isUpdatable" : "0",
@@ -72,9 +71,9 @@
       "name" : "MLModelType_mlProgram"
     },
     "userDefinedMetadata" : {
-      "com.github.apple.coremltools.version" : "8.3.0",
+      "com.github.apple.coremltools.source_dialect" : "TorchScript",
       "com.github.apple.coremltools.source" : "torch==2.10.0",
-      "com.github.apple.coremltools.source_dialect" : "TorchScript"
+      "com.github.apple.coremltools.version" : "8.3.0"
     },
     "inputSchema" : [
       {

GLiNERRelexBase.mlpackage/Data/com.apple.CoreML/model.mlmodel

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1cdbd638ca8fa62f41f0da645c784d4225d9e56eb824c2839cf371d5b5230115
-size 241691
+oid sha256:1fdfaa71f09541a45a64bbdfb4aab9a39f76c9b52832cef43544cc5c3c5d578d
+size 240629

GLiNERRelexBase.mlpackage/Manifest.json

@@ -1,18 +1,18 @@
 {
     "fileFormatVersion": "1.0.0",
     "itemInfoEntries": {
-        "96B523BA-C87B-4940-9EA0-1943A27C03A9": {
+        "0DA94A70-8C89-4269-9A42-B0FED31C9840": {
             "author": "com.apple.CoreML",
             "description": "CoreML Model Specification",
             "name": "model.mlmodel",
             "path": "com.apple.CoreML/model.mlmodel"
         },
-        "C89FC83E-0FD9-49B7-A95D-339E328C1CFA": {
+        "72F26EF3-1D90-464C-8B99-7444AE2024F1": {
             "author": "com.apple.CoreML",
             "description": "CoreML Model Weights",
             "name": "weights",
             "path": "com.apple.CoreML/weights"
         }
     },
-    "rootModelIdentifier": "96B523BA-C87B-4940-9EA0-1943A27C03A9"
+    "rootModelIdentifier": "0DA94A70-8C89-4269-9A42-B0FED31C9840"
 }

metadata.json

@@ -1,5 +1,8 @@
 {
   "format_version": 1,
+  "source_model": "knowledgator/gliner-relex-base-v1.0@e6a880049a19c5cc222a7a479c32e84b0d8cdd9a",
+  "architecture": "gliner UniEncoderTokenRelexGLiNER (deberta-v3-base, token_level BIO + all-pairs relations)",
+  "conversion": "torch.jit.trace -> coremltools mlprogram FP16, macOS15+, computeUnits=ALL",
   "artifacts": {
     "mlpackage": "GLiNERRelexBase.mlpackage",
     "mlmodelc": "GLiNERRelexBase.mlmodelc"
@@ -145,8 +148,26 @@
     "max_candidate_spans": 16,
     "unused_class_columns": "logit columns for class slots beyond the provided label count contain garbage (zero prompt embeddings) and MUST be ignored"
   },
+  "parity": {
+    "sentences": 12,
+    "compute_units": [
+      "ALL",
+      "CPU_ONLY"
+    ],
+    "max_token_logit_delta": 0.07953453063964844,
+    "max_rel_logit_delta": 0.08434486389160156,
+    "cpu_only_max_token_logit_delta": 0.23572158813476562,
+    "cpu_only_max_rel_logit_delta": 0.12303733825683594,
+    "gate": "exact decoded match at 0.4/0.7, logit delta < 0.1 (ALL) / < 0.3 (CPU_ONLY)"
+  },
+  "latency": {
+    "median_ms": 107.68287489190698,
+    "p90_ms": 108.86899987235665,
+    "min_ms": 91.64237510412931,
+    "runs": 30
+  },
   "sha256": {
-    "mlpackage_tree": "9d948eb9e9c690ee5941ab28261361dad00ce74d70f997ca35ac9e09a6e4ab0a",
-    "mlmodelc_tree": "f700158783796a7b93e366708c8467899ac6d4c7e18da567537046e1dfa43a5f"
+    "mlpackage_tree": "d888fed1afea1f09b74c55140eb7d1ef6d1af89a46b7870480662d0344b5dca1",
+    "mlmodelc_tree": "735ba7d1db0684e43bbe4c5ee71ffe0c0213a227207aab1c75ad014d2bbeedca"
   }
 }