Publish main JSONLite checkpoint and verified evaluation package

Main model weights, native inference package, one-SFT training lineage and evaluation evidence.

INFERENCE_VERIFICATION.json

@@ -1,1788 +1,28 @@
 {
   "model": "OpenSML-JSONLite-175M-Instruct",
   "checkpoint": "main",
-  "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
-  "requests": 60,
-  "raw_text_and_stop_matches": 60,
-  "correct_against_reference": 49,
-  "schema_valid": 60,
+  "date": "2026-10-20",
+  "playground60": {
+    "requests": 60,
+    "raw_text_token_stop_matches": 60,
+    "mismatches": [],
+    "complete_correct": 49,
+    "schema_valid": 60
+  },
+  "original1444": {
+    "requests": 1444,
+    "raw_text_token_stop_matches": 1444,
+    "mismatches": [],
+    "complete_correct": 1293,
+    "schema_valid": 1443
+  },
+  "weight_sha256": "25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725",
+  "status": "passed",
   "versions": {
+    "python": "3.11.13",
     "mlx": "0.32.3",
-    "numpy": "2.4.2",
-    "tokenizers": "0.22.2",
-    "jsonschema": "4.26.0"
+    "jsonschema": "4.26.0",
+    "tokenizers": "0.22.2"
   },
-  "platform": "Apple Silicon macOS, Python 3.11",
-  "outputs": [
-    {
-      "text": "{\"playlist_owner\":[\"my\"],\"playlist\":[\"Rock Gaming\"],\"entity_name\":[\"Mariem Hassan con Leyoad\"]}",
-      "value": {
-        "playlist_owner": [
-          "my"
-        ],
-        "playlist": [
-          "Rock Gaming"
-        ],
-        "entity_name": [
-          "Mariem Hassan con Leyoad"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 44,
-      "seconds": 0.17647575004957616,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-01",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"artist\":[\"Mariem Hassan\"],\"music_item\":[]}",
-      "value": {
-        "artist": [
-          "Mariem Hassan"
-        ],
-        "music_item": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 23,
-      "seconds": 0.09266262501478195,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-02",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"playlist_owner\":[\"my\"]}",
-      "value": {
-        "playlist_owner": [
-          "my"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 13,
-      "seconds": 0.05713866697624326,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-03",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"artist\":[],\"entity_name\":[],\"playlist\":[\"Digster Future Hits\"]}",
-      "value": {
-        "artist": [],
-        "entity_name": [],
-        "playlist": [
-          "Digster Future Hits"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 29,
-      "seconds": 0.11768570798449218,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-04",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"restaurant_type\":[\"brasserie\"],\"party_size_number\":[\"nine\"],\"state\":[\"AR\"]}",
-      "value": {
-        "restaurant_type": [
-          "brasserie"
-        ],
-        "party_size_number": [
-          "nine"
-        ],
-        "state": [
-          "AR"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 36,
-      "seconds": 0.13061708281747997,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-05",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"party_size_number\":[\"nine\"],\"timeRange\":[],\"sort\":[]}",
-      "value": {
-        "party_size_number": [
-          "nine"
-        ],
-        "timeRange": [],
-        "sort": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 27,
-      "seconds": 0.06810929113999009,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-06",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"sort\":[\"best\"],\"city\":[\"Wrigley\"]}",
-      "value": {
-        "sort": [
-          "best"
-        ],
-        "city": [
-          "Wrigley"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 21,
-      "seconds": 0.05377487512305379,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-07",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"timeRange\":[],\"poi\":[]}",
-      "value": {
-        "timeRange": [],
-        "poi": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 15,
-      "seconds": 0.04086962481960654,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-08",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"state\":[\"Louisiana\"],\"condition_temperature\":[\"colder\"]}",
-      "value": {
-        "state": [
-          "Louisiana"
-        ],
-        "condition_temperature": [
-          "colder"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 23,
-      "seconds": 0.05825637513771653,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-09",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"current_location\":[],\"geographic_poi\":[],\"condition_temperature\":[\"colder\"]}",
-      "value": {
-        "current_location": [],
-        "geographic_poi": [],
-        "condition_temperature": [
-          "colder"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 31,
-      "seconds": 0.11788337491452694,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-10",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"timeRange\":[\"last August\"],\"geographic_poi\":[\"Stunt Ranch Santa Monica Mountains Reserve\"],\"condition_temperature\":[\"colder\"]}",
-      "value": {
-        "timeRange": [
-          "last August"
-        ],
-        "geographic_poi": [
-          "Stunt Ranch Santa Monica Mountains Reserve"
-        ],
-        "condition_temperature": [
-          "colder"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 44,
-      "seconds": 0.17240966693498194,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-11",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"current_location\":[],\"geographic_poi\":[\"Stunt Ranch Santa Monica Mountains Reserve\"],\"state\":[]}",
-      "value": {
-        "current_location": [],
-        "geographic_poi": [
-          "Stunt Ranch Santa Monica Mountains Reserve"
-        ],
-        "state": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 34,
-      "seconds": 0.1316559170372784,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-12",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"artist\":[\"Behzad Mirkhani\"],\"music_item\":[\"theme\"],\"sort\":[\"last\"]}",
-      "value": {
-        "artist": [
-          "Behzad Mirkhani"
-        ],
-        "music_item": [
-          "theme"
-        ],
-        "sort": [
-          "last"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 38,
-      "seconds": 0.11877575004473329,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-13",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"service\":[],\"music_item\":[\"theme\"],\"playlist\":[]}",
-      "value": {
-        "service": [],
-        "music_item": [
-          "theme"
-        ],
-        "playlist": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 25,
-      "seconds": 0.06615762482397258,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-14",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"artist\":[\"Yui\"],\"music_item\":[\"track\"],\"service\":[\"Vimeo\"]}",
-      "value": {
-        "artist": [
-          "Yui"
-        ],
-        "music_item": [
-          "track"
-        ],
-        "service": [
-          "Vimeo"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 33,
-      "seconds": 0.08466579113155603,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-15",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"playlist\":[],\"sort\":[]}",
-      "value": {
-        "playlist": [],
-        "sort": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 14,
-      "seconds": 0.039065124932676554,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-16",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"rating_value\":[\"3\"],\"rating_unit\":[\"stars\"],\"object_name\":[\"Things Fall Apart\"]}",
-      "value": {
-        "rating_value": [
-          "3"
-        ],
-        "rating_unit": [
-          "stars"
-        ],
-        "object_name": [
-          "Things Fall Apart"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 35,
-      "seconds": 0.08983045816421509,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-17",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_name\":[\"Things Fall Apart\"],\"object_select\":[],\"object_type\":[]}",
-      "value": {
-        "object_name": [
-          "Things Fall Apart"
-        ],
-        "object_select": [],
-        "object_type": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 29,
-      "seconds": 0.07524858298711479,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-18",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"rating_value\":[\"1\"],\"object_name\":[]}",
-      "value": {
-        "rating_value": [
-          "1"
-        ],
-        "object_name": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 19,
-      "seconds": 0.049742958042770624,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-19",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_part_of_series_type\":[],\"object_name\":[]}",
-      "value": {
-        "object_part_of_series_type": [],
-        "object_name": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 23,
-      "seconds": 0.06354975001886487,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-20",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_name\":[\"Happiness Was Free\"]}",
-      "value": {
-        "object_name": [
-          "Happiness Was Free"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 16,
-      "seconds": 0.060178708052262664,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-21",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_type\":[],\"object_name\":[\"Happiness Was Free\"]}",
-      "value": {
-        "object_type": [],
-        "object_name": [
-          "Happiness Was Free"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 23,
-      "seconds": 0.0893012909218669,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-22",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_name\":[\"Martin Morning\"],\"object_type\":[\"saga\"]}",
-      "value": {
-        "object_name": [
-          "Martin Morning"
-        ],
-        "object_type": [
-          "saga"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 24,
-      "seconds": 0.07322416710667312,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-23",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_name\":[\"Martin Morning\"]}",
-      "value": {
-        "object_name": [
-          "Martin Morning"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 13,
-      "seconds": 0.040689832996577024,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-24",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"movie_type\":[\"animated movies\"],\"spatial_relation\":[\"in the neighborhood\"]}",
-      "value": {
-        "movie_type": [
-          "animated movies"
-        ],
-        "spatial_relation": [
-          "in the neighborhood"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 29,
-      "seconds": 0.08523791679181159,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-25",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"movie_type\":[\"animated movies\"],\"timeRange\":[],\"movie_name\":[]}",
-      "value": {
-        "movie_type": [
-          "animated movies"
-        ],
-        "timeRange": [],
-        "movie_name": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 32,
-      "seconds": 0.09332416695542634,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-26",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"movie_name\":[\"Mabel’s Dramatic Career\"],\"timeRange\":[\"twenty two o'clock\"]}",
-      "value": {
-        "movie_name": [
-          "Mabel’s Dramatic Career"
-        ],
-        "timeRange": [
-          "twenty two o'clock"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 36,
-      "seconds": 0.1403042080346495,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-27",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"timeRange\":[\"twenty two o'clock\"],\"spatial_relation\":[],\"object_location_type\":[]}",
-      "value": {
-        "timeRange": [
-          "twenty two o'clock"
-        ],
-        "spatial_relation": [],
-        "object_location_type": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 35,
-      "seconds": 0.1383882921654731,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-28",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"track\":[\"Blue Lantern\"],\"artist\":[\"Mira Sol\"],\"album\":[]}",
-      "value": {
-        "track": [
-          "Blue Lantern"
-        ],
-        "artist": [
-          "Mira Sol"
-        ],
-        "album": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 29,
-      "seconds": 0.11747579206712544,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-29",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"restaurant_name\":[\"River Moon\"],\"city\":[\"Denver\"],\"party_size_number\":[\"four\"],\"timeRange\":[\"tomorrow\"]}",
-      "value": {
-        "restaurant_name": [
-          "River Moon"
-        ],
-        "city": [
-          "Denver"
-        ],
-        "party_size_number": [
-          "four"
-        ],
-        "timeRange": [
-          "tomorrow"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 49,
-      "seconds": 0.18146875011734664,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-30",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"city\":[\"Ose\"],\"timeRange\":[\"Friday\"],\"country\":[]}",
-      "value": {
-        "city": [
-          "Ose"
-        ],
-        "timeRange": [
-          "Friday"
-        ],
-        "country": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 26,
-      "seconds": 0.10596829210408032,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-31",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"object_name\":[\"Glass Harbor\"],\"rating_value\":[\"four\"],\"best_rating\":[],\"rating_unit\":[\"stars\"]}",
-      "value": {
-        "object_name": [
-          "Glass Harbor"
-        ],
-        "rating_value": [
-          "four"
-        ],
-        "best_rating": [],
-        "rating_unit": [
-          "stars"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 42,
-      "seconds": 0.16606529196724296,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-32",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"movie_name\":[\"Silver Orchard\"],\"location_name\":[\"Maple Cinema\"],\"timeRange\":[]}",
-      "value": {
-        "movie_name": [
-          "Silver Orchard"
-        ],
-        "location_name": [
-          "Maple Cinema"
-        ],
-        "timeRange": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 37,
-      "seconds": 0.09538087481632829,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-33",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"entity_name\":[\"Night Steps\"],\"playlist\":[\"Evening Walks\"]}",
-      "value": {
-        "entity_name": [
-          "Night Steps"
-        ],
-        "playlist": [
-          "Evening Walks"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 27,
-      "seconds": 0.07212029211223125,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-34",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\"],\"album\":[]}",
-      "value": {
-        "artist": [
-          "Mira Sol",
-          "Theo Lane"
-        ],
-        "album": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 24,
-      "seconds": 0.06363512482494116,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-35",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"city\":[\"Bergen\",\"Orange\",\"Lake\"],\"country\":[]}",
-      "value": {
-        "city": [
-          "Bergen",
-          "Orange",
-          "Lake"
-        ],
-        "country": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 24,
-      "seconds": 0.06020650011487305,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-36",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"cuisine\":[],\"city\":[\"Boston\"],\"restaurant_name\":[]}",
-      "value": {
-        "cuisine": [],
-        "city": [
-          "Boston"
-        ],
-        "restaurant_name": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 26,
-      "seconds": 0.09844033396802843,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-37",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"album\":[\"Quiet Signals\"],\"artist\":[\"Mira Sol\"],\"track\":[]}",
-      "value": {
-        "album": [
-          "Quiet Signals"
-        ],
-        "artist": [
-          "Mira Sol"
-        ],
-        "track": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 30,
-      "seconds": 0.12414183304645121,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-38",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_name\":[\"Paper Bridge\"],\"rating_value\":[]}",
-      "value": {
-        "object_name": [
-          "Paper Bridge"
-        ],
-        "rating_value": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 20,
-      "seconds": 0.08667670795693994,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-39",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"country\":[],\"timeRange\":[],\"artist\":[]}",
-      "value": {
-        "country": [],
-        "timeRange": [],
-        "artist": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 20,
-      "seconds": 0.08656954090110958,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "FIELD-40",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"selected_playlist\":[\"night vibes\"],\"selected_artist\":[\"gn\"],\"selected_playlist_owner\":[]}",
-      "value": {
-        "selected_playlist": [
-          "night vibes"
-        ],
-        "selected_artist": [
-          "gn"
-        ],
-        "selected_playlist_owner": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 39,
-      "seconds": 0.1672995421104133,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:AddToPlaylist:train:80:view1:native:alias",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"playlist_owner\":null,\"artist\":\"gn\",\"playlist\":\"night vibes\"}",
-      "value": {
-        "playlist_owner": null,
-        "artist": "gn",
-        "playlist": "night vibes"
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 26,
-      "seconds": 0.10970429191365838,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:AddToPlaylist:train:80:view1:native:nullable",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"selected_restaurant_name\":[],\"selected_timeRange\":[\"Nov. the 1st\"],\"selected_sort\":[]}",
-      "value": {
-        "selected_restaurant_name": [],
-        "selected_timeRange": [
-          "Nov. the 1st"
-        ],
-        "selected_sort": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 40,
-      "seconds": 0.15424016606993973,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:BookRestaurant:train:684:view1:native:alias",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"restaurant_name\":\"mughlai restaurant\",\"timeRange\":\"Nov. the 1st\",\"sort\":null}",
-      "value": {
-        "restaurant_name": "mughlai restaurant",
-        "timeRange": "Nov. the 1st",
-        "sort": null
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 34,
-      "seconds": 0.1165991669986397,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:BookRestaurant:train:684:view1:native:nullable",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"selected_state\":[\"Louisiana\"],\"selected_condition_temperature\":[\"colder\"]}",
-      "value": {
-        "selected_state": [
-          "Louisiana"
-        ],
-        "selected_condition_temperature": [
-          "colder"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 29,
-      "seconds": 0.12024129112251103,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:GetWeather:train:109:view0:native:alias",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"state\":\"Louisiana\",\"condition_temperature\":\"colder\"}",
-      "value": {
-        "state": "Louisiana",
-        "condition_temperature": "colder"
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 19,
-      "seconds": 0.06711416691541672,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:GetWeather:train:109:view0:native:nullable",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"selected_playlist\":[],\"selected_music_item\":[\"theme\"],\"selected_service\":[]}",
-      "value": {
-        "selected_playlist": [],
-        "selected_music_item": [
-          "theme"
-        ],
-        "selected_service": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 34,
-      "seconds": 0.09712295792996883,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:PlayMusic:train:801:view1:native:alias",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"playlist\":null,\"music_item\":\"theme\",\"service\":null}",
-      "value": {
-        "playlist": null,
-        "music_item": "theme",
-        "service": null
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 23,
-      "seconds": 0.06972587504424155,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:PlayMusic:train:801:view1:native:nullable",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"selected_rating_value\":[\"2\"],\"selected_object_select\":[\"current\"]}",
-      "value": {
-        "selected_rating_value": [
-          "2"
-        ],
-        "selected_object_select": [
-          "current"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 29,
-      "seconds": 0.08707858389243484,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:RateBook:train:293:view0:native:alias",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_select\":\"current\",\"rating_value\":\"2\"}",
-      "value": {
-        "object_select": "current",
-        "rating_value": "2"
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 19,
-      "seconds": 0.08294720784761012,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:RateBook:train:293:view0:native:nullable",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"selected_object_name\":[\"Playlist: The Very Best of Dolly Parton\"]}",
-      "value": {
-        "selected_object_name": [
-          "Playlist: The Very Best of Dolly Parton"
-        ]
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 25,
-      "seconds": 0.11037545814178884,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:SearchCreativeWork:train:602:view1:native:alias",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"object_name\":\"Playlist: The Very Best of Dolly Parton\"}",
-      "value": {
-        "object_name": "Playlist: The Very Best of Dolly Parton"
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 20,
-      "seconds": 0.08364866697229445,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:SearchCreativeWork:train:602:view1:native:nullable",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"selected_movie_type\":[\"animated movies\"],\"selected_movie_name\":[],\"selected_location_name\":[]}",
-      "value": {
-        "selected_movie_type": [
-          "animated movies"
-        ],
-        "selected_movie_name": [],
-        "selected_location_name": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 42,
-      "seconds": 0.16512083308771253,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:SearchScreeningEvent:train:897:view1:native:alias",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"movie_type\":\"animated movies\",\"movie_name\":null,\"location_name\":null}",
-      "value": {
-        "movie_type": "animated movies",
-        "movie_name": null,
-        "location_name": null
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 31,
-      "seconds": 0.1260217921808362,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "native:snips:SearchScreeningEvent:train:897:view1:native:nullable",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\",\"Nora Vale\"],\"album\":[]}",
-      "value": {
-        "artist": [
-          "Mira Sol",
-          "Theo Lane",
-          "Nora Vale"
-        ],
-        "album": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 30,
-      "seconds": 0.12100991699844599,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "ARRAY-3-ARTIST",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"city\":[\"Bergen\",\"Ilmenau\",\"Trondheim\"],\"country\":[]}",
-      "value": {
-        "city": [
-          "Bergen",
-          "Ilmenau",
-          "Trondheim"
-        ],
-        "country": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 27,
-      "seconds": 0.09397912514396012,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "ARRAY-3-CITY",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\",\"Nora Vale\",\"Eli Brooks\"],\"album\":[]}",
-      "value": {
-        "artist": [
-          "Mira Sol",
-          "Theo Lane",
-          "Nora Vale",
-          "Eli Brooks"
-        ],
-        "album": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 35,
-      "seconds": 0.10037354100495577,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "ARRAY-4-ARTIST",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"city\":[\"Bergen\",\"Ilmenau\",\"Trondheim\",\"Stavanger\"],\"country\":[]}",
-      "value": {
-        "city": [
-          "Bergen",
-          "Ilmenau",
-          "Trondheim",
-          "Stavanger"
-        ],
-        "country": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 32,
-      "seconds": 0.12685562507249415,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "ARRAY-4-CITY",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    },
-    {
-      "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\",\"Nora Vale\",\"Eli Brooks\",\"Iris Reed\"],\"album\":[]}",
-      "value": {
-        "artist": [
-          "Mira Sol",
-          "Theo Lane",
-          "Nora Vale",
-          "Eli Brooks",
-          "Iris Reed"
-        ],
-        "album": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 40,
-      "seconds": 0.16119341715238988,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "ARRAY-5-ARTIST",
-      "raw_text_and_stop_match": true,
-      "exact": true
-    },
-    {
-      "text": "{\"city\":[\"Bergen\",\"Eden\",\"Tromsø\"],\"country\":[]}",
-      "value": {
-        "city": [
-          "Bergen",
-          "Eden",
-          "Tromsø"
-        ],
-        "country": []
-      },
-      "schema_valid": true,
-      "success": true,
-      "stop_reason": "turn_end",
-      "output_tokens": 26,
-      "seconds": 0.10789529187604785,
-      "output_mode": "native-json",
-      "model_calls": 1,
-      "json_assembly": false,
-      "output_repair": false,
-      "constrained": false,
-      "protocol": {
-        "temperature": 0,
-        "repetition_penalty": 1,
-        "max_tokens": 256,
-        "serialization": "native-sft-chat-v1"
-      },
-      "id": "ARRAY-5-CITY",
-      "raw_text_and_stop_match": true,
-      "exact": false
-    }
-  ],
-  "model_calls_per_request": 1,
-  "json_assembly": false,
-  "output_repair": false,
-  "constrained_decoding": false,
-  "scope": "Stock-MLX packaging parity on the observed 60 playground fixtures; not a fresh blind benchmark or full 1,444-case cross-runtime parity.",
-  "elapsed_seconds": 6.364110708935186
+  "scope": "Live native packaged-source generation; text, output-token count and stop reason match saved selected-checkpoint generations for every tested request. No repair, assembly or constrained decoding."
 }

PACKAGING_VERIFICATION.json

@@ -1,23 +1,31 @@
 {
   "model": "OpenSML-JSONLite-175M-Instruct",
-  "date": "2026-10-16",
+  "date": "2026-10-20",
   "checkpoint": "main",
   "weight_file": "model.safetensors",
   "weight_bytes": 351242342,
-  "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
-  "native_bundle_sha256": "c5bb8eaa1dd4e3bbfb094192c3b91203f20aba776eb73dc29fc47831fa043087",
+  "weight_sha256": "25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725",
+  "native_bundle_sha256": "8f5a075f63e1ff8857af3ddd916afc57742e614b1da5cbbe15452e9c84e94d2a",
+  "source_snapshot_files": 418,
   "runtime": "stock MLX 0.32.3; Apple Silicon; Python 3.11",
-  "requests": 60,
-  "raw_runtime_matches": 60,
-  "complete_correct": 49,
-  "schema_valid": 60,
-  "root_runner_verification": "source archive hash and snapshot verified; live CLI receipt saved separately",
+  "requests": 1504,
+  "raw_runtime_matches": 1504,
+  "playground60": {
+    "requests": 60,
+    "raw_text_token_stop_matches": 60,
+    "mismatches": [],
+    "complete_correct": 49,
+    "schema_valid": 60
+  },
+  "original1444": {
+    "requests": 1444,
+    "raw_text_token_stop_matches": 1444,
+    "mismatches": [],
+    "complete_correct": 1293,
+    "schema_valid": 1443
+  },
   "json_assembly": false,
   "output_repair": false,
-  "tokenizer_root_matches_bundle": true,
-  "documentation_update": {
-    "date": "2026-10-20",
-    "scope": "Intended use and limitations; updated documentation archive",
-    "model_tokenizer_and_native_inference_algorithm_unchanged": true
-  }
+  "constrained_decoding": false,
+  "tokenizer_root_matches_bundle": true
 }

README.md

@@ -21,6 +21,7 @@ datasets:
 - Avelina/python-edu-cleaned
 - rajpurkar/squad_v2
 - databricks/databricks-dolly-15k
+- AmazonScience/massive
 ---
 
 # OpenSML-JSONLite-175M-Instruct
@@ -31,7 +32,7 @@ datasets:
 
 **A small language model that generates complete JSON directly from source text and a supplied JSON Schema.** It makes one autoregressive call; parsing/schema validation follows generation. No JSON assembly, output repair, constrained decoder or secondary model is used.
 
-The model was fine-tuned for grounded question answering, exact and focused field extraction, and direct JSON generation. Its final joint JSON extraction stage combines original training replay, multiple-value examples and authored prose. Weights are provided as `model.safetensors`. The pretrained ancestor is [OpenSML-Base-175M](https://huggingface.co/OpenSML/OpenSML-Base-175M).
+The model combines structured-data continued pretraining with one joint SFT for grounded extraction and direct schema-guided JSON generation. Weights are provided as `model.safetensors`. The pretrained ancestor is [OpenSML-Base-175M](https://huggingface.co/OpenSML/OpenSML-Base-175M).
 
 ## Intended use and limitations
 
@@ -41,7 +42,7 @@ This is a research model for **extracting information that is explicitly stated
 
 **Outside the verified scope:** nested objects, arrays of objects, numeric or Boolean outputs, optional-property logic, enums, references and advanced schema constraints. General chat, multilingual input and long-document extraction are also unverified. This describes the released runner's tested contract, not a claim that the weights can never produce these formats.
 
-**Check both structure and meaning.** Parse the output, validate it against the schema, and check the values against the source. The reported 100% schema validity applies to the observed 1,444-request evaluation; complete correctness there was 88.02%. It is not a promise for new inputs. The instruction, source, schema and generated answer must share the 2,048-token context; the supplied runner allows up to 256 output tokens.
+**Check both structure and meaning.** Parse the output, validate it against the schema, and check the values against the source. The reported 99.93% schema validity applies to the observed 1,444-request evaluation; complete correctness there was 89.54%. It is not a promise for new inputs. The instruction, source, schema and generated answer must share the 2,048-token context; the supplied runner allows up to 256 output tokens.
 
 ## Run locally
 
@@ -75,9 +76,7 @@ Supply source text and a flat object JSON Schema with 1–8 required properties
 
 ## Verified source/schema/output examples
 
-Each example below is a complete walkthrough: **Prompt instruction + Source text + JSON Schema → Model output**. Open an example to see all four parts together.
-
-The instruction, source text and schema are the model's inputs. The model output is its observed response—not part of the prompt. These examples were verified with the released model; other requests can fail.
+Each example groups **Prompt instruction, Source text, JSON Schema and Model output** in one expandable section. Only the first three are model inputs.
 
 ### Track and artist
 
@@ -94,8 +93,6 @@ The instruction, source text and schema are the model's inputs. The model output
 
 #### 3. JSON Schema
 
-This defines the requested fields and their types.
-
 ```json
 {
   "type": "object",
@@ -133,7 +130,7 @@ This defines the requested fields and their types.
 
 #### 4. Model output
 
-The JSON below is generated by the model. It is shown here for reference and is not sent as an input.
+Recorded generation; the output is not supplied as input.
 
 ```json
 {
@@ -147,17 +144,12 @@ The JSON below is generated by the model. It is shown here for reference and is
 }
 ```
 
-**Try this example locally:**
-
 ```bash
 python inference.py --example 29
 ```
 
-This command supplies the instruction, source text and schema automatically.
-
 </details>
 
-
 ---
 
 ### Restaurant request
@@ -171,22 +163,20 @@ This command supplies the instruction, source text and schema automatically.
 
 #### 2. Source text
 
-> Book River Moon in Denver for four people tomorrow.
+> Find a Thai restaurant in Boston.
 
 #### 3. JSON Schema
 
-This defines the requested fields and their types.
-
 ```json
 {
   "type": "object",
   "properties": {
-    "restaurant_name": {
+    "cuisine": {
       "type": "array",
       "items": {
         "type": "string"
       },
-      "description": "restaurant name"
+      "description": "cuisine style"
     },
     "city": {
       "type": "array",
@@ -195,26 +185,18 @@ This defines the requested fields and their types.
       },
       "description": "city or town"
     },
-    "party_size_number": {
-      "type": "array",
-      "items": {
-        "type": "string"
-      },
-      "description": "explicit party size"
-    },
-    "timeRange": {
+    "restaurant_name": {
       "type": "array",
       "items": {
         "type": "string"
       },
-      "description": "date, time, or time-range expression"
+      "description": "restaurant name"
     }
   },
   "required": [
-    "restaurant_name",
+    "cuisine",
     "city",
-    "party_size_number",
-    "timeRange"
+    "restaurant_name"
   ],
   "additionalProperties": false
 }
@@ -222,36 +204,26 @@ This defines the requested fields and their types.
 
 #### 4. Model output
 
-The JSON below is generated by the model. It is shown here for reference and is not sent as an input.
+Recorded generation; the output is not supplied as input.
 
 ```json
 {
-  "restaurant_name": [
-    "River Moon"
+  "cuisine": [
+    "Thai"
   ],
   "city": [
-    "Denver"
-  ],
-  "party_size_number": [
-    "four"
+    "Boston"
   ],
-  "timeRange": [
-    "tomorrow"
-  ]
+  "restaurant_name": []
 }
 ```
 
-**Try this example locally:**
-
 ```bash
-python inference.py --example 30
+python inference.py --example 37
 ```
 
-This command supplies the instruction, source text and schema automatically.
-
 </details>
 
-
 ---
 
 ### Multiple artists
@@ -269,8 +241,6 @@ This command supplies the instruction, source text and schema automatically.
 
 #### 3. JSON Schema
 
-This defines the requested fields and their types.
-
 ```json
 {
   "type": "object",
@@ -300,7 +270,7 @@ This defines the requested fields and their types.
 
 #### 4. Model output
 
-The JSON below is generated by the model. It is shown here for reference and is not sent as an input.
+Recorded generation; the output is not supplied as input.
 
 ```json
 {
@@ -312,17 +282,12 @@ The JSON below is generated by the model. It is shown here for reference and is
 }
 ```
 
-**Try this example locally:**
-
 ```bash
 python inference.py --example 35
 ```
 
-This command supplies the instruction, source text and schema automatically.
-
 </details>
 
-
 ---
 
 ### Nullable fields
@@ -340,8 +305,6 @@ This command supplies the instruction, source text and schema automatically.
 
 #### 3. JSON Schema
 
-This defines the requested fields and their types.
-
 ```json
 {
   "$schema": "https://json-schema.org/draft/2020-12/schema",
@@ -380,7 +343,7 @@ This defines the requested fields and their types.
 
 #### 4. Model output
 
-The JSON below is generated by the model. It is shown here for reference and is not sent as an input.
+Recorded generation; the output is not supplied as input.
 
 ```json
 {
@@ -390,62 +353,62 @@ The JSON below is generated by the model. It is shown here for reference and is
 }
 ```
 
-**Try this example locally:**
-
 ```bash
 python inference.py --example 48
 ```
 
-This command supplies the instruction, source text and schema automatically.
-
 </details>
 
-
 ## Training
 
+The model starts from the from-scratch 175,611,648-parameter base at update 132005 (17,300,111,360 processed positions), then inherits **100,096,883 additional tokens of structured-data continued pretraining**, followed by **one joint full-weight SFT**. It does not inherit weights from earlier extraction SFT experiments.
 
-The base was pretrained from scratch on 17,300,111,360 token positions using five Apple Silicon Macs with Thunderbolt/RDMA distributed training. Intended mixture: 50% FineWeb-Edu, 18% DCLM-Edu, 13% English FineWiki, 9% Cosmopedia v2,10% educational Python. These are configured sampling shares, not unique-document counts. Base held-out validation loss: 2.5563359552.
+The continuation samples 25% web, 9% DCLM, 6.5% English wiki, 4.5% Cosmopedia, 5% educational Python, 5% JSON records, 15% schema/instance documents, 20% prose/schema/JSON documents and 10% Python working with JSON. These are configured token-mixture shares. Continuation uses next-token loss over all real document/EOS tokens, BF16 compute and imported FP32-master AdamW state. It warms from 1e-5 to 3.221225472e-5 over 5M additional tokens, then holds that rate. The selected continuation checkpoint contains 100,096,883 additional tokens; later continuation checkpoints are not inherited. Small fixed next-token validation panels are not extraction-accuracy benchmarks.
 
-![Training loss and held-out validation](pretraining-training-loss.png)
+The joint SFT prepares **36,000 presentations** covering original SNIPS extraction (15,000), schema variation (3,000), grounded QA rendered as JSON (3,000), official MASSIVE TRAIN slot extraction (3,000), authored prior tasks (2,400), atomic JSON tasks (3,200) and composite JSON tasks (6,400). Atomic/composite tasks cover types, missing and optional values, lists, role assignment, corrections, nested records, references, composition and constraints. Coverage in training does not establish reliable mastery of each feature.
 
-*Training loss and held-out validation during pretraining.*
+The main checkpoint is selected at update 500 of a 563-update experiment, inheriting **32,000 presentations from 21,392 distinct source texts**. Checkpoint metadata records **1,027,385 supervised assistant/EOS positions**; this is not a count of all serialized input positions. The exact selected-prefix component counts and source hashes are in `provenance/SELECTED_PREFIX.json`.
 
-The latest full-weight SFT jointly mixes original source/schema/JSON training replay, controlled multiple-value examples and authored natural-language passages across the seven SNIPS domains. The prepared corpus contains **21,144 presentations**: 15,000 original JSON views, 768 cardinality replay views and 5,376 controlled prose presentations (1,753 distinct new texts). The released main checkpoint inherits **19,200 presentations** from one deterministic shuffle: 13,640 original JSON views, 689 cardinality views and 4,871 prose views. It processes **6,153,194 serialized positions**, including **633,561 supervised assistant positions**. Later updates from the completed experiment are not included.
+SFT uses a fresh full-weight AdamW optimizer, BF16 compute with FP32 master weights/moments, batch 64, microbatch 2, betas (0.9, 0.95), epsilon 1e-8, no weight decay, clipping at 1, and seed 202610180142. The learning rate warms for 25 updates to 3e-5, then follows cosine decay toward 3e-6 over 563 updates. Only assistant JSON and EOS contribute to mean per-request token loss; source, schema and instruction tokens are masked. The training sequence limit is 1,536 within the 2,048-token model context. There is no teacher distillation, LoRA or RL in this SFT, and no teacher is needed for inference. The development evaluation selected update 500 before the broader four-suite evaluation.
 
-The run uses full-weight BF16 compute with FP32-master AdamW, batch 64, microbatch 2, betas (0.9, 0.95), epsilon 1e-8, zero weight decay and gradient clipping at 1. A fresh optimizer warms up over 20 updates to 5e-6, then follows a cosine schedule toward 1e-6 over the full 331-update experiment. Only assistant JSON and the turn-ending token contribute to mean per-request token cross-entropy. There is no teacher KL, LoRA or RL in this latest stage. Source text, schema and instruction tokens are masked. The inherited earlier cardinality stage used training-only distillation; no teacher is used at inference.
+![Original base training history](pretraining-training-loss.png)
 
-The full inherited lineage and earlier SFT data are documented in the technical report.
+*This figure describes the original base run. Continued pretraining and SFT are separately documented above.*
 
 ## Native JSON evaluation
 
-The released model was evaluated on 1,444 SNIPS-derived source-text and JSON Schema requests using raw greedy generation:
-
-| Metric | Released model |
+| Metric | Main checkpoint |
 | --- | ---: |
-| Entire request correct | **88.02%** |
-| Field accuracy | **94.17%** |
-| Extracted-value F1 | **92.88%** |
-| Extracted-value precision | **92.42%** |
-| Extracted-value recall | **93.35%** |
-| Missing-field accuracy | **95.22%** |
-| Schema-valid JSON | **100.00%** |
+| Entire request correct | **89.54%** |
+| Field accuracy | **94.91%** |
+| Extracted-value F1 | **93.75%** |
+| Extracted-value precision | **93.54%** |
+| Extracted-value recall | **93.96%** |
+| Missing-field accuracy | **96.13%** |
+| Schema-valid JSON | **99.93%** |
 
-The model answers **1,271/1,444 requests completely correctly**. Paired-source correctness is **561/694**. The evaluation records **53 false abstentions**, **73 missed missing fields**, and **24 unsupported predicted values**. All 1,444 outputs parse, validate and stop naturally; there are no output-budget failures.
+The original SNIPS-derived panel contains **1,444 requests**, of which **1,293 (89.54%)** are completely correct. All 1,444 outputs parse and stop naturally; **1,443/1,444 (99.93%)** validate against the supplied schema. One output misses a required city property. Recorded counts include 49 false abstentions, 59 missed missing fields and 20 unsupported predicted values. Valid JSON is not proof of correct extraction.
 
-The separate observed 60-case playground scores **49/60 (81.67%)**. Its original 54-case subset scores 46/54; the six added three/four/five-item probes score 3/6. Three larger-list probes fail. The compact 240-request development panel scores 178/240 (74.17%). These panels measure different scopes and must not be pooled. They have prior research exposure and are not fresh blind benchmarks or an external leaderboard.
+The original 60-case playground scores **49/60 (81.67%)**. The expanded 108-case playground scores **78/108 (72.22%)**: 49/60 original extraction cases and 29/48 added structural probes. The separate dedicated 384-case JSON diagnostic scores **183/384 (47.66%)**, with **81.51% schema validity**. These broader structural diagnostics use a different generic instruction and include schemas outside the supplied flat-schema runner contract. They demonstrate remaining weaknesses, not verified support for arbitrary JSON Schema.
 
-These aggregate scores do not establish universal retention or production reliability. Checkpoint-selection checks and remaining failures are documented in the technical report.
+The adapted English MASSIVE TEST slot-list evaluation scores **1,565/2,974 (52.62%)** complete requests, with **50.11% slot F1** and **99.73% schema validity**. Official MASSIVE TRAIN examples were included in SFT; TEST examples were evaluation-only. This custom schema/JSON rendering is not the official MASSIVE leaderboard protocol.
+
+Each panel measures a different scope; do not pool their scores or interchange structural-path F1, extracted-value F1 and slot F1. The authored diagnostics and original extraction panel have prior research exposure and are not fresh blind benchmarks. Some requests share source text or construction frames. References and expected answers are used only after generation. No answer hints, output repair, JSON assembly, secondary model or constrained decoding are used. The original panel, expanded playground and dedicated diagnostic allow 256 output tokens; adapted MASSIVE allows 384. All share a 2,048-token context and reject overflow without truncation.
 
 ![Complete-request correctness on separate observed panels](complete-request-exact.svg)
 
-## Limitations
+## Detailed limitations
+
+The intended public runner contract is short English evidence extraction into one flat closed JSON object with 1–8 required string, nullable-string or string-array fields. Dates and numbers stay as copied text. Accepted input does not guarantee correct output. The model can omit or merge entities, shorten spans, confuse roles and invent unsupported values. Unfamiliar key names, descriptions and domains require their own validation. No reliable maximum entity count has been established.
+
+Training includes broader JSON tasks, but their measured reliability remains limited. On the dedicated diagnostic, nested objects score **13/24**, role binding **5/24**, parallel groups **4/24**, numeric types **6/24**, and the missing/optional family **4/24** completely correct requests. The last number is whole-request correctness for that challenging family, not raw missing-field accuracy. Canonical and renamed schema variants can behave differently. Deep nesting, record association and optional-property decisions remain research weaknesses. The supplied flat-schema runner rejects broader schemas rather than implying verified support.
 
-A focused English extraction research model for documented flat schemas and mostly SNIPS fields. It can shorten, omit, invent or misassign values, confuse roles, miss list members, and fail on corrections or unfamiliar definitions. Valid JSON does not prove correct extraction. Long documents, nested objects, numeric/boolean outputs, multilingual tasks, general chat and prompt-injection robustness are outside the verified scope. `success` means structural validity and natural completion, not semantic correctness.
+General chat, coding, multilingual quality, long documents, calibrated confidence and prompt-injection robustness are not established. Check both structure and meaning: parse, validate the schema, then compare values and roles with source evidence. `success` reports valid structure plus natural completion, not factual correctness. Constrained-decoding scores from other checkpoints do not apply to these weights; no matched constrained/unconstrained result is claimed here.
 
-## Released model identity and package
+## Main checkpoint identity and package
 
-Selected BF16 weight SHA-256:
+BF16 weight SHA-256:
 
-`781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560`
+`25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725`
 
-Includes `model.safetensors`, tokenizer, native MLX runner, source bundle, 60 source/schema/reference examples, raw evaluation records, checksums and provenance. Standalone verification checks generated strings and stop reasons against the saved 60-case protocol; this is packaging parity, not an independent accuracy benchmark. Source uses Apache-2.0; upstream terms apply separately; no trained-weight license is assigned by this release.
+Includes weights, the frozen tokenizer, native MLX runner, 60 source/schema/reference examples, recorded generations, four-suite evaluation evidence, training source and provenance. The package uses one raw greedy generation per request. Source uses Apache-2.0; data and dependencies retain upstream terms. No trained-weight license is assigned. See [technical report](TECHNICAL_REPORT.md) and [run guide](RUN_GUIDE.md).

RUN_GUIDE.md

@@ -14,7 +14,7 @@ This is a research model for **extracting information that is explicitly stated
 
 **Outside the verified scope:** nested objects, arrays of objects, numeric or Boolean outputs, optional-property logic, enums, references and advanced schema constraints. General chat, multilingual input and long-document extraction are also unverified. This describes the released runner's tested contract, not a claim that the weights can never produce these formats.
 
-**Check both structure and meaning.** Parse the output, validate it against the schema, and check the values against the source. The reported 100% schema validity applies to the observed 1,444-request evaluation; complete correctness there was 88.02%. It is not a promise for new inputs. The instruction, source, schema and generated answer must share the 2,048-token context; the supplied runner allows up to 256 output tokens.
+**Check both structure and meaning.** Parse the output, validate it against the schema, and check the values against the source. The reported 99.93% schema validity applies to the observed 1,444-request evaluation; complete correctness there was 89.54%. It is not a promise for new inputs. The instruction, source, schema and generated answer must share the 2,048-token context; the supplied runner allows up to 256 output tokens.
 
 ```bash
 python3 -m venv .venv
@@ -70,11 +70,11 @@ python native_bundle/inference.py --weights "$PWD/model.safetensors" --example 2
 ```
 
 Selected weights:
-`781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560`.
+`25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725`.
 The current source module is inference_native/native_json.py. The old labeled
 adapter, if present in the historical source archive, is not used by this
 entry point. Optimizer state and the original training workspace are not needed
 for inference. See the technical report for lineage, metrics and limitations.
 
 
-The main checkpoint adds joint source/schema/JSON SFT with replay, cardinality and authored prose. No teacher is needed at inference. Exact lineage and settings are in checkpoint_provenance.json.
+The main checkpoint uses structured continued pretraining followed by one joint extraction/JSON SFT. The public flat-schema runner contract remains as described above. Training broader schemas does not establish reliable broader-schema inference. See the technical report for exact provenance and separate diagnostics.

SHA256SUMS

@@ -1,21 +1,21 @@
 5d659853ddccca76e4090ce3e4ebe15bc6017239752a54d663f5e997a695651a  .gitattributes
 4a49b666643257b6a7f8bc5e7051a4a1389cdd021778d14a736b2f5f1a2c005d  CITATION.cff
-9a4fb5f559d82ed0120a57139eb1e1c616bfdb18c39cca7b1e9bcab24ef968bc  INFERENCE_VERIFICATION.json
-74a074f90bd5e7a26d5f6cd5757ddc1fbfcc72e259e0b23a210fa48d91b35fdb  JSONLITE_RESULTS.json
+e2e6e75848a42ea4bb5e50ab85a91f1e1cd2b413327fb9667d4893e15a86db04  INFERENCE_VERIFICATION.json
+b6da2e34e1d59693c11b257203804260613f636f40d5fca56d4ae8b1e253ae3e  JSONLITE_RESULTS.json
 c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4  LICENSE
-7e092bbf3d2e770493e1e31cee471e04f59f2b26355d4ded86565e3e39bc8ade  PACKAGING_VERIFICATION.json
-5e828c09db086de986dad8eafebebeb6ce21185934a6e8e6236cab049741ff2d  README.md
-e54ba0ed507dd5da8a6fe54cd420f91f0db1bed31e75fc8ff5934fe883f638bc  RUN_GUIDE.md
-c6191baf937bce61f1b747acfbcb975a17eef6e8a598682a751a1031d1cfee23  SOURCE_NOTICE.md
-7bb05614604bf206027f8897409959dae8fbd67e417addc63da55d8f3eac3fb8  TECHNICAL_REPORT.md
-e4e44da050abbd930748ec91731ddf1be16be83a4aa59bd8c4fc63478b384a03  THIRD_PARTY_NOTICES.md
-663087d32872824f7f2fec425ad4054f308913c43da7c3f83521c7011967fdbe  checkpoint_metadata.json
-a365a7697b463bacc6e44e26d11a1e62d43d96b5d52c70c2e9df8d4513f86537  checkpoint_provenance.json
-ecec6f191822d16a3da0b571fa129f5016e1243ca7d567de476a9d0e8ad66646  complete-request-exact.svg
-981e02e7777980705770cacf19f16d76ea74e85e7edb018f875ddb4766afe19d  config.json
-6c8e504441f93b0d1ed428bc0cd7973788b7be8b7b7983d47ad20ce8b7a1ddeb  inference.py
-781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560  model.safetensors
-c5bb8eaa1dd4e3bbfb094192c3b91203f20aba776eb73dc29fc47831fa043087  native_bundle.zip
+bc70f2792409ca4af51a61b3925b9de0690dffc19155bb168ca5b983552c35db  PACKAGING_VERIFICATION.json
+75d1d6e30f98bf99c8112e6a33e1f6405db6df3a808ef6ab404f14ba3abc3181  README.md
+82a05c21b027d9cebd4d512c284c309159385dcebe457172bc9a350c74483a71  RUN_GUIDE.md
+a0e232a2b134e771355f018706c69bf4fa46ab5c3c65f1c94314d5f0acb73e80  SOURCE_NOTICE.md
+5fdd5f40546b36c59a1ea84f5eeac64b0b7b36cb133affa5dfad53ff8d0c4d5b  TECHNICAL_REPORT.md
+d79d1b4947327762b6833bdd147949c15fd2a60bf65865442e7ba0fcfd67a4a8  THIRD_PARTY_NOTICES.md
+3118935bf70f5872d9f0b4342999f1aad29f457288a893ca43d9c7cc798c76b1  checkpoint_metadata.json
+edb05bb77440adc61f2833ee92af2abb2ccebd18bed22be66b2d4d2626461eec  checkpoint_provenance.json
+369cab2772baaa6dacc26dbceeb66cb3cc7df1aac49aec3fc747ce418908a35f  complete-request-exact.svg
+778bed0cc245b238e1f08ab3f67ddb2dc5cda79095c65b598437fb4fcf9c8a27  config.json
+0957a6347cb5052847b713bfb8048e7aa2c84b79b6cc1080b0c64f2d3b7f73a9  inference.py
+25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725  model.safetensors
+8f5a075f63e1ff8857af3ddd916afc57742e614b1da5cbbe15452e9c84e94d2a  native_bundle.zip
 641fe860d6fd24547fe4d15bd852190f00ddec31bd74eb7ca9eeae74031c2f0c  pretraining-training-loss.png
 c7b1a3c0efb09ef3b5416a24072e48b25e2cc1f122fb5ed638c0f95b72594af9  requirements.txt
 e095f0413a11fa91fbb58f37d00e4c6bcda59beb89b28f6f680db1436a7e8891  tokenizer.json

SOURCE_NOTICE.md

@@ -1,26 +1,3 @@
-# Source and data notice
+# Source and artifact scope
 
-Current inherited weights: Stage A/132005 → SFT-065/469 → SFT-068/379 →
-SFT-093/160 → SFT-100/469 → additional extraction SFT (main checkpoint). The final stage directly supervises native JSON.
-Other research trials and unused helper code are not weight ancestors.
-Author-controlled source uses Apache-2.0. Upstream datasets/dependencies retain
-their notices; no trained-weight license is assigned by this source update.
-GitHub excludes weight and optimizer binaries; the HF companion has the weights.
-
-Pretraining uses the pinned educational web/wiki/Cosmopedia/Python mixture.
-Grounded QA uses selected SQuAD2/Dolly and checked authored fictional facts.
-SNIPS original CC0 chunk annotations supply exact extraction and native JSON
-labels. The 15,000 curated JSON records are source/schema/target views of 12,115
-TRAIN requests, not 15,000 distinct documents. Development/test are source-disjoint
-from TRAIN within this curated dataset, but have previous research exposure.
-Pretraining decontamination and perfectly correct semantic golds are not claimed.
-
-Raw pretraining pools, packed arrays, optimizer state, credentials, SSH keys and
-private host/network receipts are excluded. Published metadata replaces private
-workspace paths with [workspace]; original and published hashes are separated.
-60 actual local fixtures, raw predictions and held-out evidence are retained.
-The canonical held-out 1,388 and schema-stress 56 requests are included as JSONL in
-evaluation/native-json-heldout.jsonl and the prepared native data directory.
-See THIRD_PARTY_NOTICES.md, docs/TECHNICAL_REPORT.md and provenance/LINEAGE.json.
-
-The latest stage jointly replays original SNIPS TRAIN JSON views, controlled cardinality views and authored English prose with original TRAIN values. It adds 19,200 inherited presentations; no teacher KL is used in this stage. Prepared authored prose and targets are included for inspection, with observed evaluation scope and remaining failures in the report.
+The native inference package contains the model architecture, tokenizer, fixed prompt, raw decoder, validation and 60 source/schema/reference fixtures. Frozen training recipes, prepared SFT data/order, selected checkpoint metadata and evaluation records are included for inspection. No optimizer binaries, parent weights or packed pretraining pools are distributed. Training code is archival and uses the original workspace conventions; create local configuration and readiness checks before attempting reproduction. The public runner is a native MLX implementation.

TECHNICAL_REPORT.md

@@ -1,18 +1,12 @@
 # OpenSML-JSONLite-175M-Instruct: native JSON technical report
 
-**Author:** William Zebrowski · **Parameters:** 175,611,648 · **Date:** 2026-10-16
+Date: 2026-10-20. Main checkpoint: BF16 `model.safetensors`, 175,611,648 parameters, 351,242,342 bytes.
 
-## 1. Current artifact and behavior
+Weight SHA-256: `25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725`.
 
-The main checkpoint directly generates JSON from source text and a supplied JSON Schema. A single native autoregressive request emits the entire object. There is no fieldwise extraction-to-JSON assembler, answer repair, constrained decoder or gold-conditioned filling in the current inference path. Parsing and schema validation happen only after generation; the raw generated string is retained even when incorrect or invalid.
+## 1. Intended use and limitations
 
-Current artifact: **main checkpoint**, BF16 `model.safetensors`, 351,242,342 bytes.
-SHA-256: `781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560`.
-This is a custom native MLX architecture, not a Transformers/MLX-LM auto-loader export.
 
-This report supersedes the earlier fieldwise SFT-093 release. Its historical report is retained as `HISTORICAL_SFT093_REPORT.md`; those adapter-assisted scores are not current native JSON scores. No release-readiness or arbitrary-schema reliability claim is made.
-
-### Intended use in plain language
 
 This is a research model for **extracting information that is explicitly stated in short English text** and placing it into one flat JSON object. Supply a schema with clear field descriptions. The verified runner accepts 1–8 required fields containing strings, nullable strings or arrays of strings. Dates and numbers are copied as text rather than normalized or converted.
 
@@ -20,26 +14,22 @@ This is a research model for **extracting information that is explicitly stated
 
 **Outside the verified scope:** nested objects, arrays of objects, numeric or Boolean outputs, optional-property logic, enums, references and advanced schema constraints. General chat, multilingual input and long-document extraction are also unverified. This describes the released runner's tested contract, not a claim that the weights can never produce these formats.
 
-**Check both structure and meaning.** Parse the output, validate it against the schema, and check the values against the source. The reported 100% schema validity applies to the observed 1,444-request evaluation; complete correctness there was 88.02%. It is not a promise for new inputs. The instruction, source, schema and generated answer must share the 2,048-token context; the supplied runner allows up to 256 output tokens.
+**Check both structure and meaning.** Parse the output, validate it against the schema, and check the values against the source. The reported 99.93% schema validity applies to the observed 1,444-request evaluation; complete correctness there was 89.54%. It is not a promise for new inputs. The instruction, source, schema and generated answer must share the 2,048-token context; the supplied runner allows up to 256 output tokens.
 
-## 2. Exact inherited lineage
 
-`Stage A → grounded QA → requested-field extraction → focused extraction → native JSON foundation → cardinality/replay extraction → joint JSON extraction SFT`
+## 2. Exact inherited lineage
 
-Only those selected checkpoints are inherited. Other trials and failed RL/JSON experiments are research history, not hidden additional ancestors. The architecture and frozen tokenizer remain unchanged. Each SFT stage resets its local counters and AdamW optimizer while continuing the inherited neural weights.
+| Stage | Selected checkpoint | Inherited training |
+| --- | --- | --- |
+| Original from-scratch pretraining | 132005 | 17,300,111,360 processed positions |
+| Structured continued pretraining | 132838 | 100,096,883 additional real tokens |
+| One joint extraction/JSON SFT | 500 of 563 | 32,000 presentations; 1,027,385 supervised assistant/EOS positions |
 
-| Stage | Inherited update | Presentations through selected artifact | Serialized positions |
-| --- | ---: | ---: | ---: |
-| Stage A pretrained | 132005 | Document stream | 17,300,111,360 lifetime pretraining positions |
-| SFT-065 grounded QA | 469 of 782 | 15,008 | 3,097,331 |
-| SFT-068 requested-field extraction | 379 | 12,115 | 1,842,593 |
-| SFT-093 focused extraction/replay | 160 | 5,120 | 706,399 |
-| SFT-100 native JSON | 469 | 15,000 | 4,634,126 |
-| Additional extraction SFT | 64 selected of 96 completed | 2,048 | 600,558 |
-| Joint JSON extraction SFT | 300 of 331 | 19,200 | 6,153,194 |
-| Inherited post-training total | 1,841 updates | 68,491 | 17,034,201 |
+Original base weight SHA-256: `5151567e266244c9e32fd6e01775192633abe227320269ecdefe28ce43e9ccc3`.
+Continued-pretraining parent weight SHA-256: `57d0933d5d806e7f589a60170b5080ff46895b732be480aa953650701f4e19cc`.
+Main weight SHA-256: `25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725`.
 
-The latest stage metadata counter records 633,561 supervised assistant positions; its serialized count above was recomputed from the exact selected shuffle prefix. Presentation counts include repeated source views. Serialized positions include masked prompt tokens; they are not supervised-output-token counts. The earlier 150M model weights did not initialize this model; the earlier frozen OpenSML tokenizer was reused.
+The main weights inherit one SFT. Earlier SFT research recipes informed the data design but their weights and optimizers are not ancestors of this artifact. All other continued-pretraining endpoints are outside this selected lineage. Supervised SFT counters and pretraining token counters measure different things and must not be added as if equivalent.
 
 ## 3. Architecture and parameterization
 
@@ -172,134 +162,80 @@ The 175.6M-parameter base was trained from random initialization using next-toke
 *Training loss and held-out validation during pretraining. Left: raw training loss and exponential moving average. Right: held-out validation loss.*
 
 
-## 7. Inherited supervised foundation and prior extraction continuation
+## 7. Continued pretraining and joint SFT
 
-The four inherited foundation stages remain unchanged: grounded evidence QA (15,008 presentations through its selected checkpoint), requested-field SNIPS extraction (12,115), focused single-field extraction/replay(5,120), then directly supervised native JSON (15,000). Their original source and checkpoint records remain in provenance. Each stage uses full model weights, fresh AdamW, BF16 computation with FP32 master weights/moments/gradient accumulation, and assistant-output-only supervision. No RL, LoRA or added extraction head is in this lineage.
+The model starts from the from-scratch 175,611,648-parameter base at update 132005 (17,300,111,360 processed positions), then inherits **100,096,883 additional tokens of structured-data continued pretraining**, followed by **one joint full-weight SFT**. It does not inherit weights from earlier extraction SFT experiments.
 
-The native JSON foundation has 12,115 distinct SNIPS TRAIN requests, with 12,115 canonical string-array schemas, 1,443 renamed-key views and 1,442 nullable-string views. Source text, field definitions and the complete schema are input; complete JSON and stop tokens are the assistant target. Original CC0 SNIPS chunk annotations provide source spans. Metadata fidelity and schema validation do not establish perfect semantic annotation. Frozen tokenizer and architecture are unchanged.
+The continuation samples 25% web, 9% DCLM, 6.5% English wiki, 4.5% Cosmopedia, 5% educational Python, 5% JSON records, 15% schema/instance documents, 20% prose/schema/JSON documents and 10% Python working with JSON. These are configured token-mixture shares. Continuation uses next-token loss over all real document/EOS tokens, BF16 compute and imported FP32-master AdamW state. It warms from 1e-5 to 3.221225472e-5 over 5M additional tokens, then holds that rate. The selected continuation checkpoint contains 100,096,883 additional tokens; later continuation checkpoints are not inherited. Small fixed next-token validation panels are not extraction-accuracy benchmarks.
 
-### Prior cardinality/replay extraction SFT
+The joint SFT prepares **36,000 presentations** covering original SNIPS extraction (15,000), schema variation (3,000), grounded QA rendered as JSON (3,000), official MASSIVE TRAIN slot extraction (3,000), authored prior tasks (2,400), atomic JSON tasks (3,200) and composite JSON tasks (6,400). Atomic/composite tasks cover types, missing and optional values, lists, role assignment, corrections, nested records, references, composition and constraints. Coverage in training does not establish reliable mastery of each feature.
 
-This completed continuation starts from the native JSON foundation. It adds matched zero/one/two/three-value supervision for artists and cities, including zero-primary-value cases where an album or country is explicitly present. It retains original JSON TRAIN replay and adds frozen-parent distribution distillation on replay assistant tokens. Every model batch has 24 replay and 8 new views. This is an inherited ancestor of the current main checkpoint.
+The main checkpoint is selected at update 500 of a 563-update experiment, inheriting **32,000 presentations from 21,392 distinct source texts**. Checkpoint metadata records **1,027,385 supervised assistant/EOS positions**; this is not a count of all serialized input positions. The exact selected-prefix component counts and source hashes are in `provenance/SELECTED_PREFIX.json`.
 
-| Setting | Value |
-|---|---|
-| Prepared corpus |3,072 presentations: 2,304 original TRAIN replay +768 authored schema views|
-| Distinct authored source requests in prepared corpus |256|
-| Prepared replay variants |1,344 canonical, 480 renamed, 480 nullable|
-| Completed experiment |96 optimizer updates, one pass, 900,453 serialized positions|
-| Inherited selected prefix |2,048 presentations, 600,558 positions, 64 updates|
-| Inherited prefix mixture |1,536 replay views from 1,482 source texts +512 authored views from 250 source texts|
-| Model adaptation |Full weights; Mac 1 only; fresh AdamW|
-| Batch / microbatch |32 /2|
-| Peak / final schedule LR |2e-6 /2e-7 over 96 updates|
-| Warmup |6 updates; cosine afterward|
-| Betas / epsilon |(0.9,0.95) /1e-8|
-| Weight decay / clipping |0.01 /1.0|
-| Seed |20261119|
+SFT uses a fresh full-weight AdamW optimizer, BF16 compute with FP32 master weights/moments, batch 64, microbatch 2, betas (0.9, 0.95), epsilon 1e-8, no weight decay, clipping at 1, and seed 202610180142. The learning rate warms for 25 updates to 3e-5, then follows cosine decay toward 3e-6 over 563 updates. Only assistant JSON and EOS contribute to mean per-request token loss; source, schema and instruction tokens are masked. The training sequence limit is 1,536 within the 2,048-token model context. There is no teacher distillation, LoRA or RL in this SFT, and no teacher is needed for inference. The development evaluation selected update 500 before the broader four-suite evaluation.
 
-The current weights inherit only this ancestor’s selected prefix; updates after that ancestor are not part of these weights. The endpoint's actual learning rate is recorded by the cosine schedule, not the planned final LR. Presentation counts include multiple schemas for the same text and do not mean distinct user queries.
+### Selected SFT prefix
 
-For every row, CE is averaged over supervised assistant tokens within each sequence, then averaged across sequences. On original TRAIN replay add 1.0×KL(frozen foundation || student), using the full 32,000-token distribution, temperature 1 and teacher-forced ground-truth histories. Only assistant JSON and turn/document-stop positions participate. The reference network is frozen/stop-gradient; new authored rows use CE alone. This regularizer is training-only: no teacher is needed for inference, and it does not guarantee retention or repair parent errors.
+| Component | Presentations |
+| --- | ---: |
+| atomic | 2,844 |
+| composite | 5,689 |
+| grounded_qa | 2,667 |
+| massive | 2,667 |
+| prior_authored | 2,133 |
+| schema_variation | 2,667 |
+| snips | 13,333 |
 
-The new views use literal original SNIPS TRAIN values with split-disjoint authored value pools/phrases, excluding observed evaluation source texts and holding new authored DEV/TEST values out of replay. Source text, exact offsets, schema/gold agreement, masks, token budgets and hashes were checked. Twenty-eight tests, GPU accumulation checks, frozen-reference checks and a temporary optimizer/checkpoint-resume smoke passed. Authored examples are programmatic controlled supervision, not independently collected user traffic. SGD, MIT Movie and other newly researched datasets are **not** in this checkpoint.
+## 8. Inference protocol
 
+The native runner accepts the displayed fixed extraction instruction, source text and schema only. It emits a complete JSON object in one autoregressive call, with greedy temperature 0 and repetition penalty 1. It performs no constrained decoding, assembly or repair. Validation happens afterward. Source, instruction, schema and generated answer share a 2,048-token context; overflow is rejected without truncation. Public-runner output budget is 1–256 tokens. The runner supports the documented flat, closed string-based schema contract, not the entire JSON Schema standard. The tokenizer is unchanged. Strict parsing rejects duplicate keys and nonfinite numbers.
 
-### Latest joint JSON extraction SFT
+## 9. Evaluation
 
-The latest full-weight SFT jointly mixes original source/schema/JSON training replay, controlled multiple-value examples and authored natural-language passages across the seven SNIPS domains. The prepared corpus contains **21,144 presentations**: 15,000 original JSON views, 768 cardinality replay views and 5,376 controlled prose presentations (1,753 distinct new texts). The released main checkpoint inherits **19,200 presentations** from one deterministic shuffle: 13,640 original JSON views, 689 cardinality views and 4,871 prose views. It processes **6,153,194 serialized positions**, including **633,561 supervised assistant positions**. Later updates from the completed experiment are not included.
+| Metric | Main checkpoint |
+| --- | ---: |
+| Entire request correct | **89.54%** |
+| Field accuracy | **94.91%** |
+| Extracted-value F1 | **93.75%** |
+| Extracted-value precision | **93.54%** |
+| Extracted-value recall | **93.96%** |
+| Missing-field accuracy | **96.13%** |
+| Schema-valid JSON | **99.93%** |
 
-The run uses full-weight BF16 compute with FP32-master AdamW, batch 64, microbatch 2, betas (0.9, 0.95), epsilon 1e-8, zero weight decay and gradient clipping at 1. A fresh optimizer warms up over 20 updates to 5e-6, then follows a cosine schedule toward 1e-6 over the full 331-update experiment. Only assistant JSON and the turn-ending token contribute to mean per-request token cross-entropy. There is no teacher KL, LoRA or RL in this latest stage. Source text, schema and instruction tokens are masked. The inherited earlier cardinality stage used training-only distillation; no teacher is used at inference.
+The original SNIPS-derived panel contains **1,444 requests**, of which **1,293 (89.54%)** are completely correct. All 1,444 outputs parse and stop naturally; **1,443/1,444 (99.93%)** validate against the supplied schema. One output misses a required city property. Recorded counts include 49 false abstentions, 59 missed missing fields and 20 unsupported predicted values. Valid JSON is not proof of correct extraction.
 
-## 8. Prompt, schema and native inference
+The original 60-case playground scores **49/60 (81.67%)**. The expanded 108-case playground scores **78/108 (72.22%)**: 49/60 original extraction cases and 29/48 added structural probes. The separate dedicated 384-case JSON diagnostic scores **183/384 (47.66%)**, with **81.51% schema validity**. These broader structural diagnostics use a different generic instruction and include schemas outside the supplied flat-schema runner contract. They demonstrate remaining weaknesses, not verified support for arbitrary JSON Schema.
 
-Supply source text and a flat closed JSON Schema with 1–8 required fields. Supported properties are strings, nullable strings and arrays of strings. The requested behavior is to return every required key, with [] for absent arrays and null for absent nullable strings; generated answers can still be incorrect. Copy case/spelling and complete values; preserve source order. Nested schemas, numbers/booleans, enums and advanced constraints are outside the verified runner contract. The supplied runner rejects overlong inputs without truncation.
+The adapted English MASSIVE TEST slot-list evaluation scores **1,565/2,974 (52.62%)** complete requests, with **50.11% slot F1** and **99.73% schema validity**. Official MASSIVE TRAIN examples were included in SFT; TEST examples were evaluation-only. This custom schema/JSON rendering is not the official MASSIVE leaderboard protocol.
 
-The fixed instruction is followed by compact `{"source_text":...,"schema":...}` and native SFT role tokens. It issues exactly one autoregressive whole-object generation with temperature 0, repetition penalty 1 and at most 256 output tokens. No fieldwise assembly, repair, schema-constrained decoder, normalization, reference answer, frozen teacher or secondary model is used at inference. Parsing/schema checks only annotate the raw generated text afterward.
+Each panel measures a different scope; do not pool their scores or interchange structural-path F1, extracted-value F1 and slot F1. The authored diagnostics and original extraction panel have prior research exposure and are not fresh blind benchmarks. Some requests share source text or construction frames. References and expected answers are used only after generation. No answer hints, output repair, JSON assembly, secondary model or constrained decoding are used. The original panel, expanded playground and dedicated diagnostic allow 256 output tokens; adapted MASSIVE allows 384. All share a 2,048-token context and reject overflow without truncation.
 
-## 9. Evaluation and release selection
+### Original extraction correctness by domain
 
-| Metric | Main checkpoint | Previous release |
+| Domain | Correct / requests | Rate |
 | --- | ---: | ---: |
-| Entire request correct | **88.02%** | 83.03% |
-| Field accuracy | **94.17%** | 91.36% |
-| Extracted-value F1 | **92.88%** | 90.40% |
-| Extracted-value precision | **92.42%** | 90.53% |
-| Extracted-value recall | **93.35%** | 90.26% |
-| Missing-field accuracy | **95.22%** | 92.79% |
-| Schema-valid JSON | **100.00%** | 99.79% |
-
-Complete correctness is **1,271/1,444**, compared with 1,199/1,444 for the previous release: **96 improved requests and 24 regressions**, a net gain of 72. All seven domain aggregate correctness totals improve. Paired-source correctness rises from 512/694 to **561/694**. False abstentions fall from 107 to **53**, missed missing fields from 110 to **73**, and unsupported predicted values from 27 to **24**. All 1,444 outputs parse, validate and stop naturally; there are no output-budget failures.
-
-The separate observed 60-case playground scores **49/60 (81.67%)**, compared with 45/60. Its original 54-case subset scores 46/54; the six added three/four/five-item probes score 3/6. Three larger-list probes still fail. The compact 240-request development panel scores 178/240 (74.17%). These panels measure different scopes and must not be pooled. They have prior research exposure and are not fresh blind benchmarks or an external leaderboard.
-
-The main checkpoint was selected for aggregate improvement despite failing the strict zero-lost-case development gate. Individual regressions remain; improved aggregate scores do not establish universal retention or production reliability.
-
-| Metric | Main checkpoint | Previous release |
-| --- | ---: | ---: |
-| Entire request correct | **88.02%** | 83.03% |
-| Field accuracy | **94.17%** | 91.36% |
-| Extracted-value F1 | **92.88%** | 90.40% |
-| Extracted-value precision | **92.42%** | 90.53% |
-| Extracted-value recall | **93.35%** | 90.26% |
-| Missing-field accuracy | **95.22%** | 92.79% |
-| Schema-valid JSON | **100.00%** | 99.79% |
-
-### Correctness by domain
-
-| Domain | Previous | Main | Requests |
-| --- | ---: | ---: | ---: |
-| AddToPlaylist | 165 | 172 | 206 |
-| BookRestaurant | 162 | 172 | 208 |
-| GetWeather | 164 | 178 | 206 |
-| PlayMusic | 153 | 164 | 208 |
-| RateBook | 193 | 195 | 204 |
-| SearchCreativeWork | 197 | 199 | 208 |
-| SearchScreeningEvent | 165 | 191 | 204 |
-
-### Protocol and contamination limits
-
-Greedy temperature 0, repetition penalty 1, maximum 256 output tokens and the unchanged native serialization. Every request supplies only the source text, schema and fixed extraction instruction. References, gold spans and offsets are used after generation only. No prompt truncation, constrained decoding, secondary model, JSON assembly or repair is used. Strict parsing rejects duplicate keys and nonfinite constants. Complete correctness requires a natural stop, valid JSON and schema, and exact agreement of all fields. Lists retain order; value precision/recall/F1 compare extracted values, while field accuracy includes missing fields. The full machine-readable result includes counts, denominators, paired correctness, unsupported values and stopping metrics.
-
-The 1,444 requests comprise the original canonical and schema-stress views; some views reuse the same source. Previously observed tests and playground examples informed research and owner selection. They are not a fresh blind holdout. Synthetic prose shares construction families and training entities, so gains do not establish open-domain generalization. The selected checkpoint is a preserved no-pruning replay artifact identified by its exact hash; metrics are recomputed from that artifact's saved generations, not borrowed from an earlier artifact. No claim of bitwise training replay identity is made.
+| AddToPlaylist | 179/206 | 86.89% |
+| BookRestaurant | 173/208 | 83.17% |
+| GetWeather | 186/206 | 90.29% |
+| PlayMusic | 169/208 | 81.25% |
+| RateBook | 199/204 | 97.55% |
+| SearchCreativeWork | 200/208 | 96.15% |
+| SearchScreeningEvent | 187/204 | 91.67% |
 
 ## 10. Failure audit and limitations
 
-### What the released model is intended to do
-
-The intended task is short English text extraction into a supplied, flat object schema: copy stated values into the requested fields, keep separate entities as separate string-array items, and use empty arrays or nullable strings for absent information where the schema allows it. Clear field descriptions matter. The runner accepts 1–8 required fields with strings, nullable strings or arrays of strings; dates and numeric expressions remain strings. Those input limits describe the runner contract and do not guarantee that every accepted request will be answered correctly.
-
-### What can still go wrong
-
-The model can shorten names or titles, omit or merge list members, confuse roles or field meanings, and fill a missing field with an unsupported value. Corrections, unfamiliar field names, new domains and complex associations require application-specific evaluation. No tested field count or entity count establishes a reliable extraction maximum. Optional-property logic, nesting, arrays of records, numeric/Boolean outputs, enums, references and advanced constraints are outside the released runner's verified contract. Root-level arrays or primitives, general chat, multilingual use and long-document extraction are not supported claims of this release. Unverified does not mean impossible for the weights; it means the supplied runner and release evidence do not establish dependable support.
-
-| Evidence for this released checkpoint | What it establishes | What it does not establish |
-| --- | --- | --- |
-| 1,271/1,444 complete requests correct (88.02%) | Exact whole-answer performance on the documented, previously observed SNIPS-derived panel | Equivalent accuracy on arbitrary schemas or new domains |
-| 1,444/1,444 schema-valid outputs | All outputs met the tested flat schemas in that evaluation | Correct extracted values, or universal schema compliance |
-| 49/60 complete playground requests correct (81.67%) | Performance on a separate observed diagnostic panel | A fresh blind benchmark or a general production reliability rate |
-| 3/6 complete three/four/five-item probes correct | Larger lists remain inconsistent | A guaranteed maximum number of extractable entities |
+The intended public runner contract is short English evidence extraction into one flat closed JSON object with 1–8 required string, nullable-string or string-array fields. Dates and numbers stay as copied text. Accepted input does not guarantee correct output. The model can omit or merge entities, shorten spans, confuse roles and invent unsupported values. Unfamiliar key names, descriptions and domains require their own validation. No reliable maximum entity count has been established.
 
-These measurements use the released checkpoint identified in section 1. Local research checkpoints are separate artifacts; their broader-schema or per-family scores are not measurements of this release. The panels have prior research exposure and must not be pooled into one accuracy number.
+Training includes broader JSON tasks, but their measured reliability remains limited. On the dedicated diagnostic, nested objects score **13/24**, role binding **5/24**, parallel groups **4/24**, numeric types **6/24**, and the missing/optional family **4/24** completely correct requests. The last number is whole-request correctness for that challenging family, not raw missing-field accuracy. Canonical and renamed schema variants can behave differently. Deep nesting, record association and optional-property decisions remain research weaknesses. The supplied flat-schema runner rejects broader schemas rather than implying verified support.
 
-### Checks required in an application
+General chat, coding, multilingual quality, long documents, calibrated confidence and prompt-injection robustness are not established. Check both structure and meaning: parse, validate the schema, then compare values and roles with source evidence. `success` reports valid structure plus natural completion, not factual correctness. Constrained-decoding scores from other checkpoints do not apply to these weights; no matched constrained/unconstrained result is claimed here.
 
-Parse the raw output, validate it against the supplied schema, then check that extracted values are supported by the source and assigned to the right fields. A valid string can still name the wrong person, omit an item or invent a value; schema validation cannot detect all such errors. The runner's `success` flag means structural validity and natural completion, not semantic correctness. The full instruction, source, schema and answer share a 2,048-token context, and the released runner allows up to 256 generated tokens. Overlong inputs are rejected rather than truncated. Validate the complete workflow on representative application data before relying on its outputs.
+## 11. Packaging and reproducibility
 
-### Additional failure-audit observations
+Checkpoint identifiers, original and sanitized source hashes, the exact SFT data/order and selected prefix, continued-pretraining recipe, and saved generated answers are included. Optimizer binaries, packed corpus pools and parent model weights are not bundled. Reproduction needs the exact parent weight hashes, upstream data under its own terms, compatible dependencies and fresh local preflight checks. Published sanitized receipts are audit evidence, not resumable optimizer state; absolute workspace references in archival training code need local configuration. The supported self-contained runtime is `inference.py`, not a Transformers or MLX-LM auto-loader.
 
-There are 173 entirely incorrect original-test requests and 24 previously correct requests become incorrect. Errors include incomplete copying, wrong roles, missing or extra values and unsupported values. All seven domain totals improve, but per-request retention is not perfect. On the six longer-list probes only three are correct. Flat schema validity is 100% on this test, not a guarantee for arbitrary prompts, schema types or novel domains. No general chat, reasoning, code, multilingual, long-document or safety readiness claim is made. The authored development panel is small and repeatedly observed.
-
-## 11. Packaging, verification and reproduction
-
-The main package uses BF16 weights, the unchanged tokenizer and the supplied native MLX runner. The verification receipt records stock-MLX raw-string and stop-reason parity on all 60 observed playground fixtures. This establishes packaging parity within the stated scope, not full 1,444-case cross-runtime parity or an independent benchmark. Standard `model.safetensors` and the public model ID remain unchanged. Previous Hub and Git revisions retain rollback identities.
-
-The source archive includes the frozen latest training recipe, prepared JSONL corpus and input hashes, exact prefix counts, saved generated answers, scorer and matched parent comparisons. Public copies sanitize private workspace paths and distinguish original from distributed hashes. Optimizer binaries and packed pretraining pools are omitted. Reproduction requires the specified parent and layout, compatible dependencies, and fresh local preflight receipts; sanitized audit receipts are not resumable optimizer state. See TRAINING_AND_REPRODUCIBILITY.md, provenance/LINEAGE.json, results/JSONLITE_RESULTS.json and pretraining/native_json_128/ in the archive.
+The source archive contains a SHA-256 snapshot manifest. `INFERENCE_VERIFICATION.json` records the live public-runner comparison against saved main-checkpoint generations. `results/JSONLITE_RESULTS.json` separates each evaluation panel and protocol. `provenance/LINEAGE.json`, `provenance/SELECTED_PREFIX.json`, and `TRAINING_AND_REPRODUCIBILITY.md` document training identity. Evaluation correctness is checked after independent generation; references never enter the inference prompt.
 
 ## 12. Attribution and licensing
 
-
-
-Source code retains Apache-2.0; upstream data and dependencies retain their notices. Original SNIPS annotations are CC0; the matched cardinality examples are programmatically authored from TRAIN literals. The earlier SQuAD/Dolly/pretraining source notices remain in THIRD_PARTY_NOTICES.md. Source licensing does not automatically assign a trained-weight license; none is assigned by this update.
-
-Hugging Face edition: archive-relative code/evidence references resolve inside native_bundle.zip.
+See `THIRD_PARTY_NOTICES.md`, `licenses/` and the included source manifests for pinned upstream records. Original pretraining uses FineWeb-Edu, DCLM-Edu, English FineWiki, Cosmopedia and educational Python. SFT reuses annotated official TRAIN material and authored tasks, including SNIPS and MASSIVE, plus grounded-QA-derived rows. Source, upstream datasets and trained weights have separate licensing status. Apache-2.0 applies to author-controlled source contributions; no trained-weight license is assigned by this release.

THIRD_PARTY_NOTICES.md

@@ -19,6 +19,8 @@ No IFEval, Paraloq, SOB, ARC, PIQA, HellaSwag or other experiment's scores are a
 
 For actual source-file pins and identity-preserving adaptations, see the technical report and `provenance/`. These declarations are attribution records, not a blanket commercial-use or model-weight redistribution determination. The public weight license remains unassigned.
 
-## Latest joint JSON extraction data
+## Joint SFT and structured continuation
 
-The latest stage reuses SNIPS TRAIN source/schema/JSON views and authored controlled prose with original TRAIN values. Original SNIPS attribution and CC0 terms above continue to apply. Authored wording does not make evaluation independent or establish a new weight license. No SGD/MIT dataset was added to this stage.
+The SFT uses official SNIPS TRAIN and MASSIVE en-US TRAIN annotations plus authored/schema-transformed tasks and inherited grounded-QA-derived text rows. MASSIVE: Amazon Science MASSIVE contributors, CC-BY-4.0; https://huggingface.co/datasets/AmazonScience/massive . The included prepared rows and their provenance identify original sources and transformations. SQuAD and Dolly declarations above apply to their derived prepared rows where present; prepared derived text is distributed here with its recorded upstream attribution. TEST annotations are used only for evaluation. Authored facts and schema frames are correlated synthetic research material.
+
+Structured continuation adds authored JSON records, schema/instance documents, prose/schema/JSON documents and Python/JSON code alongside the original corpus mixture. It is next-token training, not benchmark-answer supervision. Dataset use does not assign a blanket trained-weight license.

checkpoint_metadata.json

@@ -1,24 +1,44 @@
 {
-  "joint128": {
+  "recipe": {
+    "model": {
+      "attention_impl": "fast",
+      "bias": false,
+      "ce_impl": "metal",
+      "d_model": 768,
+      "ffn_impl": "packed-metal",
+      "loss_dtype": "float32",
+      "max_seq_len": 2048,
+      "mlp_multiple_of": 256,
+      "mlp_ratio": 4.0,
+      "n_heads": 12,
+      "n_kv_heads": 4,
+      "n_layers": 24,
+      "qk_norm": true,
+      "rope_base": 10000.0,
+      "vocab_size": 32000
+    }
+  },
+  "restart130": {
     "contract": {
+      "arm": "100m",
       "files": {
-        "pretraining/data/sft_128_joint_extraction_20261016/manifest.json": "5172c7702b2c0b6fa9583052007acb69630bed6b23b5fb4d3ce62d3666ca1a00",
-        "pretraining/json_sampling_119/common.py": "b3ef439b09b8ca21d9bb454a9afe377f202ea75bda522e51e7b284399b39d0c0",
-        "pretraining/json_sampling_119/data.py": "5e560b14b795a58f2ca2cbe6d7fb19924b8ec633788cdef8875d80449e3734a4",
-        "pretraining/json_sampling_119/run.py": "84fe9caab3c196fec5e71c56a1c4ab8e7c63309fe4529811680d72e4a5cec747",
-        "pretraining/native_json_128/__init__.py": "fcdbfacd29cfb9f3235160fe00d4c9f4f4c35cb27d469f1e6f3796f137b373b2",
-        "pretraining/native_json_128/common.py": "354cda3dd100321fa858e64ae22e4ecc27e09d97f2c824811e3b9f53f205be52",
-        "pretraining/native_json_128/data.py": "874ab8eae8281b1967ea7ed22a0bad122e0c830a7ee8391ae9d201e0d6867087",
-        "pretraining/native_json_128/evaluate.py": "2dc61a8269c7c279ca13814f7e0b07ddbee477308ac5abc87d1461ca75cfedc8",
-        "pretraining/native_json_128/runtime.py": "5a83f2ba3f900f69390d10c2a67d10cd2d7654a1b22345f059a02488b0239ef8",
-        "pretraining/native_json_128/scoring.py": "1f4d2c69ebc640f993b77cb09ecf280d2df1dbf3f20d1060f7c8a5ce6780dd97",
-        "pretraining/native_json_128/train.py": "dded6b80814d12f621eafe0ddcc41481fa80cc98a2182be442fb09c31a5735e7",
-        "pretraining/native_json_128/verify.py": "738c709a8358a0818f6dbd42d81a9db25ea76569cededae345e1a0d474344779",
-        "pretraining/runs/sft_105_cardinality_distillation_20261014/step_0000064_cdd8be1e8d5a/model.safetensors.json": "8b8c0049d731f8124eaec78a213d05c24a520bc4f190dc82975d763f43351bbc",
-        "pretraining/scripts/run_sft_128_joint_extraction.sh": "dcf97394bf39afd46c5591bcc19abd8802422d4ab3e5d9fd8845c5e46f493ead",
+        "pretraining/data/sft_152_cpt_comparison_20261019/manifest.json": "ea818a53da76ab187f211c4ded3c97c1d7a59c2d5775d2043bda67753d2d5539",
+        "pretraining/json_cpt_compare_152/__init__.py": "1d3ce2fd36e9fcbef1d5d578b9f7842cb672309966a317c89e8ac01a23b00636",
+        "pretraining/json_cpt_compare_152/common.py": "4b347223dd9b6e83d67c5d2f8ec48ad6e82f2dff97c3e59427e8f6864bf84e22",
+        "pretraining/json_cpt_compare_152/preflight.py": "8703ac7786bf5b7518a5ec15d333adeb6bfa41c84ae5c1406b1661dac8783dd9",
+        "pretraining/json_cpt_compare_152/prepare.py": "678fa106aaafb7e344a917d4f4e85101d8cf11234e6ff8b933f0449443ab5807",
+        "pretraining/json_cpt_compare_152/run.py": "80de9d542992ef52313fe4bb02d43dc6b8d98f16204a039f04161ac0a3b5c06d",
+        "pretraining/json_foundation_142/common.py": "6f6727d92f879bfc5eb6f060e87c3423a4cf7a1b22f1249c31f5c937ba8ab34d",
+        "pretraining/json_foundation_142/run.py": "23027ddf8e0f16a0d3e2aaaae0c3324c2efad7190cbe492fe9818f7dcdb310f5",
+        "pretraining/json_foundation_142/runtime.py": "f562c4d159853ce1291949de92b179a2cb08e6fbd781476e6732e1f0b4a842a9",
+        "pretraining/json_foundation_142/scoring.py": "a3136e24be36bcc3ef0fe886d4a02000bb63c2ccafab157b566353533259056b",
+        "pretraining/json_restart_130/common.py": "748737c19bb9a6d087528f46270c63cbd1c0c4193023b18be92ca43857f7fe6e",
+        "pretraining/json_restart_130/runtime.py": "5a3826b2124f1eea2832a4d753a227c7f5f6c62bf882ffdb800cd6e20aeb6ce2",
+        "pretraining/json_restart_130/train.py": "5d9b7a163dad5557a016eb59fc739d91e2ec342e26bab81e894e90a75336f7ea",
+        "pretraining/jsonlite_eval/common.py": "f4514b8ff0e6861d051c5e24d20358912e9800f91957fead70f53ee34cfb078c",
+        "pretraining/runs/cpt_151_structured_20261018/step_0132838_224f03f16825/model.safetensors.json": "05cb4f0bdbabf4398afae42930eac11efd8e765360e0f5ed04cc3d2fbf78c0dd",
+        "pretraining/scripts/run_sft_152_cpt_comparison.sh": "0301fbca595fb546069a580d447a6fb4e60bbbf8a3801eaf6893a483e1d3716c",
         "pretraining/sft/data.py": "70348a9df1473da88330695d5409f17b929b9bc1dc79f77525272f6d87317aa1",
-        "pretraining/sft/evaluate.py": "80624551512112850145ccb57d6435d7a1bae86d0b95f8f52d56b8cb6f102a76",
-        "pretraining/sft/quality.py": "2abf5926a22a33b4f29c699855b24f7e459b6e631aa6ef780876ddc205f17758",
         "pretraining/sml_v1/checkpoint_bundle.py": "73cbf0c17cf74a44bb7c74a8f9e5bb63671057a1f664a6a4af0e0777b2c8f1a4",
         "pretraining/sml_v1/common.py": "ec4dda551cbe45cbd30d64c7c82379a8849ce02a3ad1a597a293844b5dc86351",
         "pretraining/sml_v1/fast_loss.py": "f99edeedb6cb328ef728d05643f65ea20371437f7e5806009ae3ad0885bafbf1",
@@ -26,19 +46,16 @@
         "pretraining/sml_v1/model.py": "1eee6f8afe78767ff6b9c604ccf2328c98b24bc94dbad6291a0ad699dd7345fe",
         "pretraining/sml_v1/precision.py": "b9f535fca724153703ed57a300eceb5b34a8f8381fc65bf5c4c5a0f7e066303c",
         "pretraining/sml_v1/tokenization.py": "254e1f5d60cc4498a9f1209955752db28b91839f48d3121d674f3f5ac42e52e5",
-        "pretraining/tests/test_native_json_128.py": "7547e08bf1f06d87bdcb34197b412ae9555c9e3a503b3d608d21c27e9f423c3c",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/corpus.json": "e4c47b7b6aba08f948de7c2c7cb94c58dbd09dde2ded6ac185c60856ee686d27",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/manifest.json": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/report.json": "e871f690eabebf82da173f0e406f38b79a10cd7d2647bb3592c3fc35ae9a2f16",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/tokenizer.json": "e095f0413a11fa91fbb58f37d00e4c6bcda59beb89b28f6f680db1436a7e8891",
-        "reference/opensml-150m/tokenizer/bytebpe32k_v1/tokenizer_config.json": "619f500553367eb76f5be00a98625130ec857b9c367561334984096c029ba2a9",
-        "reports/model-design/SFT-105_cardinality_distillation_20261014/original-test-1444-step64.identity.json": "e72aeb0100d8fd145096cc8df88b20c23a4630a3c2e4b3114fa9935f10dbc96b",
-        "reports/model-design/SFT-105_cardinality_distillation_20261014/original-test-1444-step64.json": "2c730e69196aa4177ccc7bf1b09bde7a044597b7f2afde108344c9c4579a0c44",
-        "reports/model-design/SFT-128_joint_extraction_20261016/source-validation.json": "b37530d0ef3716a15e977a949a31dcca953bdb5abf32513885e8285b4f810314"
+        "reference/opensml-150m/tokenizer/bytebpe32k_v1/tokenizer_config.json": "619f500553367eb76f5be00a98625130ec857b9c367561334984096c029ba2a9"
       },
-      "parent_sha256": "655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1",
+      "format": "sft152-matched-cpt-v1",
+      "parent_sha256": "57d0933d5d806e7f589a60170b5080ff46895b732be480aa953650701f4e19cc",
       "plan": {
-        "adaptation": "all_parameters",
+        "adaptation": "full weights",
         "batch_size": 64,
         "betas": [
           0.9,
@@ -49,52 +66,41 @@
         "epochs": 1,
         "eps": 1e-08,
         "eval_every": 100,
-        "final_lr": 1e-06,
-        "max_tokens": 256,
+        "eval_steps": [
+          100,
+          200,
+          300,
+          400,
+          500,
+          563
+        ],
+        "final_lr": 3e-06,
+        "max_tokens": 384,
         "microbatch": 2,
-        "objective": "Mean per-request assistant-JSON token cross entropy, including turn EOS. No teacher KL, no inference repair or assembly.",
-        "order": "One deterministic joint shuffle, all prepared records exactly once; no late domain-only phase.",
-        "peak_lr": 5e-06,
+        "objective": "Per-request assistant JSON and EOS loss; source/schema tokens masked.",
+        "peak_lr": 3e-05,
         "precision": "BF16 compute / FP32 master AdamW",
-        "seed": 202610160128,
-        "warmup": 20,
+        "presentations": 36000,
+        "retention": "Keep every checkpoint. Matched142 control; no automatic promotion or reserved generation.",
+        "seed": 202610180142,
+        "train_max_sequence": 1536,
+        "updates": 563,
+        "warmup": 25,
         "weight_decay": 0.0
       },
+      "python": "3.11.13",
       "versions": {
         "jsonschema": "4.26.0",
         "mlx": "0.32.3",
-        "mlx-metal": "0.32.3",
         "numpy": "2.4.2",
         "tokenizers": "0.22.2"
       }
     },
-    "order_hash": "8d3c7c9d4a1e34303d2f6c6b411f50f8cc72a72f45a3dd6a75c7a60f88715be3",
-    "records": 19200,
-    "total": 331
-  },
-  "preservation_policy": {
-    "pruning": false,
-    "user_requested": true
-  },
-  "recipe": {
-    "model": {
-      "attention_impl": "fast",
-      "bias": false,
-      "ce_impl": "metal",
-      "d_model": 768,
-      "ffn_impl": "packed-metal",
-      "loss_dtype": "float32",
-      "max_seq_len": 2048,
-      "mlp_multiple_of": 256,
-      "mlp_ratio": 4.0,
-      "n_heads": 12,
-      "n_kv_heads": 4,
-      "n_layers": 24,
-      "qk_norm": true,
-      "rope_base": 10000.0,
-      "vocab_size": 32000
-    }
+    "order_hash": "e1c4dbbac134d870765f878ea3d767e6a8201325b2bcaf79fb5725fa318714f9",
+    "records": 32000,
+    "refine": false,
+    "total": 563
   },
-  "step": 300,
-  "tokens": 633561
+  "step": 500,
+  "tokens": 1027385
 }

checkpoint_provenance.json

@@ -1,83 +1,29 @@
 {
   "model": "OpenSML-JSONLite-175M",
+  "checkpoint": "main",
+  "selected_stage": "joint foundation JSON SFT",
+  "selected_update": 500,
   "stages": [
     {
-      "stage": "stage-a",
+      "stage": "original-pretraining",
       "step": 132005,
-      "tokens": 17300111360,
-      "weight_sha256": "5151567e266244c9e32fd6e01775192633abe227320269ecdefe28ce43e9ccc3",
-      "weight_bytes": 351242342,
-      "original_metadata_sha256": "15a0e3b511b97503a611c0da838a7fde1c06ebed34c7103f7cda1782cc6378a4",
-      "published_metadata_sha256": "15a0e3b511b97503a611c0da838a7fde1c06ebed34c7103f7cda1782cc6378a4"
+      "processed_positions": 17300111360,
+      "weight_sha256": "5151567e266244c9e32fd6e01775192633abe227320269ecdefe28ce43e9ccc3"
     },
     {
-      "stage": "sft-065",
-      "step": 469,
-      "tokens": 3097331,
-      "weight_sha256": "0991b805a878c15733c909ef7f4127baa2b8866f8ca7d7db232ad39434f80bc4",
-      "weight_bytes": 351242342,
-      "original_metadata_sha256": "f245e46ecfbbc499d7727c556ea889a7ac77f619d8d815f65267100ddb38e991",
-      "published_metadata_sha256": "1172eb192546ae0d4d395943be0628c8de0fa2d9a4de1243eec8e73fef2d5f44"
+      "stage": "structured-continued-pretraining",
+      "step": 132838,
+      "additional_tokens": 100096883,
+      "weight_sha256": "57d0933d5d806e7f589a60170b5080ff46895b732be480aa953650701f4e19cc"
     },
     {
-      "stage": "sft-068",
-      "step": 379,
-      "tokens": 1842593,
-      "weight_sha256": "18f0f2b8f57905ac758f392d0eb83e51e508da6a955b9f7991338e690205df86",
-      "weight_bytes": 351242342,
-      "original_metadata_sha256": "49acc0640e80749fb143e9a4d5024d2ffba7e79385314a792a46cc23f02283ac",
-      "published_metadata_sha256": "192399fe9cbb09ff0a2db1b21d87193968cc766f177dc9ad3975f289e8121989"
-    },
-    {
-      "stage": "sft-093",
-      "step": 160,
-      "tokens": 706399,
-      "weight_sha256": "d552c1380c71c7c0569e9d692ec9c46c5c94bb87a7b024e56df0f0e36df4d798",
-      "weight_bytes": 351242342,
-      "original_metadata_sha256": "32a380899c455449db066ef317de9dd8e80f6977bfb42a6fe03b67c5f23963f8",
-      "published_metadata_sha256": "32a380899c455449db066ef317de9dd8e80f6977bfb42a6fe03b67c5f23963f8"
-    },
-    {
-      "stage": "sft-100",
-      "step": 469,
-      "tokens": 4634126,
-      "records": 15000,
-      "weight_sha256": "656f45015aab548465d93bed0b57e687497163741885da7364872851fa17e9a4",
-      "weight_bytes": 351242342,
-      "original_metadata_sha256": "3864a0eaa43015d286e4da9327b0d3cea4fa4f5b80f336baabe2a549379e2e36",
-      "published_metadata_sha256": "5ecf9a8850aef6cc75add6bacda880648e72de1b38089d585e237bbce828c2c3"
-    },
-    {
-      "stage": "sft-105",
-      "public_stage": "additional extraction SFT",
-      "step": 64,
-      "tokens": 600558,
-      "records": 2048,
-      "weight_sha256": "655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1",
-      "weight_bytes": 351242342,
-      "original_metadata_sha256": "8b8c0049d731f8124eaec78a213d05c24a520bc4f190dc82975d763f43351bbc",
-      "published_metadata_sha256": "49d177fb4606673f789fbf06c694643c44504672a9f0c3f75f7eae776cb4590c"
-    },
-    {
-      "stage": "sft-128",
-      "public_stage": "joint JSON extraction SFT",
-      "step": 300,
-      "records": 19200,
-      "serialized_positions": 6153194,
-      "supervised_assistant_positions": 633561,
-      "token_counter": "supervised assistant positions; differs from earlier serialized counters",
-      "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
-      "weight_bytes": 351242342,
-      "original_metadata_sha256": "663087d32872824f7f2fec425ad4054f308913c43da7c3f83521c7011967fdbe",
-      "published_metadata_sha256": "663087d32872824f7f2fec425ad4054f308913c43da7c3f83521c7011967fdbe"
+      "stage": "joint-foundation-sft",
+      "step": 500,
+      "presentations": 32000,
+      "supervised_assistant_positions": 1027385,
+      "weight_sha256": "25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725"
     }
   ],
-  "weights_in_git": false,
-  "public_weight_location": "https://huggingface.co/OpenSML/OpenSML-JSONLite-175M",
-  "model_weight_license": "not yet assigned",
-  "scope": "Selected lineage: Stage A -> grounded QA -> requested-field extraction -> focused extraction -> native JSON foundation -> cardinality/replay distillation -> joint JSON extraction SFT. Later updates are not inherited.",
-  "selected_stage": "joint JSON extraction SFT",
-  "selected_update": 300,
-  "weight_location": "https://huggingface.co/OpenSML/OpenSML-JSONLite-175M",
-  "public_checkpoint_name": "main"
+  "sft_stages": 1,
+  "teachers_at_inference": false
 }

config.json

@@ -18,7 +18,7 @@
     "rope_base": 10000.0,
     "vocab_size": 32000
   },
-  "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
+  "weight_sha256": "25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725",
   "serialization": "native-sft-chat-v1",
   "output_mode": "native-json",
   "context_limit": 2048,
@@ -29,8 +29,8 @@
   "json_assembly": false,
   "output_repair": false,
   "constrained_decoding": false,
-  "selected_stage": "joint JSON extraction SFT",
-  "selected_update": 300,
+  "selected_stage": "joint foundation JSON SFT",
+  "selected_update": 500,
   "checkpoint_name": "main",
   "inference_teacher_required": false
 }

inference.py

@@ -12,7 +12,7 @@ import tempfile
 import zipfile
 
 ROOT = Path(__file__).resolve().parent
-BUNDLE_SHA256 = "c5bb8eaa1dd4e3bbfb094192c3b91203f20aba776eb73dc29fc47831fa043087"
+BUNDLE_SHA256 = "8f5a075f63e1ff8857af3ddd916afc57742e614b1da5cbbe15452e9c84e94d2a"
 
 def main():
     archive = ROOT / "native_bundle.zip"

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560
+oid sha256:25b406a7b86e3e8ac4dd0b7f32e27590d9d11349a0148a7815c32304150ac725
 size 351242342

native_bundle.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c5bb8eaa1dd4e3bbfb094192c3b91203f20aba776eb73dc29fc47831fa043087
-size 12835333
+oid sha256:8f5a075f63e1ff8857af3ddd916afc57742e614b1da5cbbe15452e9c84e94d2a
+size 18095722