Qwen90M step 80000; complete resumable checkpoint

complete.json

@@ -1,18 +1,18 @@
 {
-  "step": 75000,
+  "step": 80000,
   "files": {
     "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
     "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
-    "model.safetensors": "b90a19c4b2d7929e861313731bcb5beb014b3e36b39789a0beda7568ce6a846a",
+    "model.safetensors": "016fae2e58ad750ac0b4abde700bbdbea9199273ac137f7e182d0fe7874da0dc",
     "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
     "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
     "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
     "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
-    "trainer.pt": "5026bd4bb1b3b18f5fc726c01908ba56660bc127713bb70b791880fb27548eeb",
-    "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf",
+    "trainer.pt": "63c9c8c99731fc20714a54a44377517fe0f1f3043beea6e7f6b8a0c370badb1e",
+    "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783",
     "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f",
-    "shard_scheduler.json": "a365a3ff77a3205ac0d3c410749b4fb4670618f5a7de9659698fe71360225cc6",
-    "coverage.json": "2dd27baaf4f3a0abac91b8368d3639578f98d0308ab7fb49afd5e4690f165c2f",
+    "shard_scheduler.json": "ea08386ca957ff57a36e47f5f3cdc24c23475683922434b0de7b9c8abf2744b1",
+    "coverage.json": "42736094b47b18e0027534773acb49639d163b78174444895fdb7d300580166a",
     "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
   },
   "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"

coverage.json

@@ -1,107 +1,107 @@
 {
-  "step": 75000,
+  "step": 80000,
   "stats": {
-    "climbmix/blocks": 2700000,
-    "climbmix/targets": 4511757125,
-    "climbmix/input_tokens": 4512042972,
-    "ultra_qa/blocks": 1012500,
-    "ultra_qa/targets": 1651705822,
-    "ultra_qa/input_tokens": 1651709076,
-    "cortex/blocks": 1012500,
-    "cortex/targets": 38520571,
-    "cortex/input_tokens": 1609425947,
-    "ultra_style/blocks": 1012500,
-    "ultra_style/targets": 1601103718,
-    "ultra_style/input_tokens": 1601187526,
-    "openmath/blocks": 675000,
-    "openmath/targets": 1043094503,
-    "openmath/input_tokens": 1043442386,
-    "rewrite6/blocks": 337500,
-    "rewrite6/targets": 553275354,
-    "rewrite6/input_tokens": 553881803
+    "climbmix/blocks": 2880000,
+    "climbmix/targets": 4812534066,
+    "climbmix/input_tokens": 4812838946,
+    "ultra_qa/blocks": 1080000,
+    "ultra_qa/targets": 1761872361,
+    "ultra_qa/input_tokens": 1761875805,
+    "cortex/blocks": 1080000,
+    "cortex/targets": 41089673,
+    "cortex/input_tokens": 1716618096,
+    "ultra_style/blocks": 1080000,
+    "ultra_style/targets": 1707744164,
+    "ultra_style/input_tokens": 1707833604,
+    "openmath/blocks": 720000,
+    "openmath/targets": 1112712435,
+    "openmath/input_tokens": 1113083661,
+    "rewrite6/blocks": 360000,
+    "rewrite6/targets": 590150425,
+    "rewrite6/input_tokens": 590797964
   },
   "sources": {
     "climbmix": {
-      "page": 369,
-      "row": 3716,
+      "page": 394,
+      "row": 2302,
       "start": {
-        "file": 17,
+        "file": 19,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 69,
+          "row_group": 3,
           "row": 1024
         }
       },
       "epoch": 0,
-      "offset": 0,
-      "documents": 1515140
+      "offset": 2048,
+      "documents": 1616126
     },
     "rewrite6": {
-      "page": 10,
-      "row": 1413,
+      "page": 24,
+      "row": 2442,
       "start": {
         "file": 0,
         "reader": {
           "format": "jsonl-row-v1",
-          "row": 40960
+          "row": 98304
         }
       },
       "epoch": 4,
       "offset": 2048,
-      "documents": 863989
+      "documents": 922362
     },
     "openmath": {
-      "page": 231,
-      "row": 700,
+      "page": 246,
+      "row": 2574,
       "start": {
         "file": 2,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 72,
-          "row": 876
+          "row_group": 134,
+          "row": 316
         }
       },
       "epoch": 0,
       "offset": 0,
-      "documents": 946876
+      "documents": 1010190
     },
     "ultra_qa": {
-      "page": 99,
-      "row": 4047,
+      "page": 106,
+      "row": 2578,
       "start": {
         "file": 0,
         "reader": {
           "format": "parquet-row-v1",
           "row_group": 6,
-          "row": 17989
+          "row": 46661
         }
       },
       "epoch": 0,
-      "offset": 2048,
-      "documents": 409551
+      "offset": 4096,
+      "documents": 436754
     },
     "ultra_style": {
-      "page": 152,
-      "row": 2035,
+      "page": 162,
+      "row": 2876,
       "start": {
         "file": 0,
         "reader": {
           "format": "parquet-row-v1",
           "row_group": 7,
-          "row": 27468
+          "row": 68428
         }
       },
       "epoch": 0,
       "offset": 2048,
-      "documents": 624627
+      "documents": 666428
     },
     "cortex": {
-      "page": 19633,
-      "row": 27,
+      "page": 20940,
+      "row": 66,
       "start": null,
       "epoch": 0,
       "offset": 0,
-      "documents": 2513051
+      "documents": 2680386
     }
   }
 }

latest.json

@@ -1 +1 @@
-{"step": 75000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "b90a19c4b2d7929e861313731bcb5beb014b3e36b39789a0beda7568ce6a846a", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "5026bd4bb1b3b18f5fc726c01908ba56660bc127713bb70b791880fb27548eeb", "training_config.json": "2122ad6f8105987f9b87dd55d13b632084853908566b081c829ffe801259e2bf", "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f", "shard_scheduler.json": "a365a3ff77a3205ac0d3c410749b4fb4670618f5a7de9659698fe71360225cc6", "coverage.json": "2dd27baaf4f3a0abac91b8368d3639578f98d0308ab7fb49afd5e4690f165c2f", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000075000"}
+{"step": 80000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "016fae2e58ad750ac0b4abde700bbdbea9199273ac137f7e182d0fe7874da0dc", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "63c9c8c99731fc20714a54a44377517fe0f1f3043beea6e7f6b8a0c370badb1e", "training_config.json": "95ba9e002790d11959d1191b36ed994bfa10a6da51c22c772e143bec517a9783", "performance.json": "d14cb208816158eb276e72d732c3b9e6358b2a75509678412180fc272350709f", "shard_scheduler.json": "ea08386ca957ff57a36e47f5f3cdc24c23475683922434b0de7b9c8abf2744b1", "coverage.json": "42736094b47b18e0027534773acb49639d163b78174444895fdb7d300580166a", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000080000"}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b90a19c4b2d7929e861313731bcb5beb014b3e36b39789a0beda7568ce6a846a
+oid sha256:016fae2e58ad750ac0b4abde700bbdbea9199273ac137f7e182d0fe7874da0dc
 size 360622744

shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74395.06001898335,
-      "required_pages_per_second_upper_estimate": 0.00644837524988969
+      "required_token_positions_per_second": 75133.13345379168,
+      "required_pages_per_second_upper_estimate": 0.006512349584588858
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9299.382502372919,
-      "required_pages_per_second_upper_estimate": 0.003709662718355241
+      "required_token_positions_per_second": 9391.64168172396,
+      "required_pages_per_second_upper_estimate": 0.003746466284396027
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18598.765004745837,
-      "required_pages_per_second_upper_estimate": 0.004302496956763474
+      "required_token_positions_per_second": 18783.28336344792,
+      "required_pages_per_second_upper_estimate": 0.004345182031637001
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27898.147507118756,
-      "required_pages_per_second_upper_estimate": 0.0016988508129156528
+      "required_token_positions_per_second": 28174.925045171884,
+      "required_pages_per_second_upper_estimate": 0.0017157051128435733
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27898.147507118756,
-      "required_pages_per_second_upper_estimate": 0.0021476766221130506
+      "required_token_positions_per_second": 28174.925045171884,
+      "required_pages_per_second_upper_estimate": 0.0021689837231616427
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27898.147507118756,
-      "required_pages_per_second_upper_estimate": 0.34640162294496635
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28174.925045171884,
+      "required_pages_per_second_upper_estimate": 0.34983827365275444
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,412 +155,412 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13118641220319205,
-    "producer_p95_seconds": 0.39197794900246663,
-    "gpu_update_seconds": 0.9910335441786977,
-    "producer_batches_per_second": 7.62274067264771,
-    "gpu_batches_per_second": 1.009047580552617,
-    "producer_headroom": 7.554391705169171,
+    "producer_batch_seconds": 0.12603624823447035,
+    "producer_p95_seconds": 0.43727277699508704,
+    "gpu_update_seconds": 0.9812980852894166,
+    "producer_batches_per_second": 7.934225383634551,
+    "gpu_batches_per_second": 1.0190583422009507,
+    "producer_headroom": 7.785840177215272,
     "sustainable": true,
-    "buffer_seconds": 60.45304619490056,
+    "buffer_seconds": 60.84048128794383,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 443816.2390829497,
-    "upload_bytes_per_second": 5580829.247452082,
+    "generation_bytes_per_second": 287303.50039527926,
+    "upload_bytes_per_second": 6345416.31829655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.2326333939206,
+    "estimated_full_shards_per_hour": 61.64864310163291,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
     {
-      "seconds": 1.1758935670004576,
-      "bytes": 7725078,
-      "pages": 38,
+      "seconds": 0.9040151730005164,
+      "bytes": 8396980,
+      "pages": 35,
       "shards": 3
     },
     {
-      "seconds": 1.0130331639993528,
-      "bytes": 839053,
-      "pages": 26,
-      "shards": 1
-    },
-    {
-      "seconds": 1.1233831270001247,
-      "bytes": 6290212,
-      "pages": 36,
+      "seconds": 1.1415953719988465,
+      "bytes": 5808415,
+      "pages": 28,
       "shards": 2
     },
     {
-      "seconds": 0.8791882089972205,
-      "bytes": 6587099,
-      "pages": 27,
+      "seconds": 1.0399577659991337,
+      "bytes": 7078524,
+      "pages": 40,
       "shards": 3
     },
     {
-      "seconds": 1.0528832150012022,
-      "bytes": 11948549,
-      "pages": 44,
-      "shards": 3
+      "seconds": 0.9194039719986904,
+      "bytes": 5850558,
+      "pages": 28,
+      "shards": 2
     },
     {
-      "seconds": 1.0517111139997724,
-      "bytes": 6085549,
+      "seconds": 0.8959461580016068,
+      "bytes": 6085583,
       "pages": 36,
-      "shards": 2
+      "shards": 3
     },
     {
-      "seconds": 0.9597967700028676,
-      "bytes": 2343290,
-      "pages": 28,
+      "seconds": 0.9613766149996081,
+      "bytes": 2539274,
+      "pages": 33,
       "shards": 2
     },
     {
-      "seconds": 1.0771120079989487,
-      "bytes": 10854426,
-      "pages": 39,
+      "seconds": 1.3802653160018963,
+      "bytes": 12066636,
+      "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 1.2815551900021092,
-      "bytes": 7356201,
+      "seconds": 0.828995023999596,
+      "bytes": 4743589,
+      "pages": 35,
+      "shards": 2
+    },
+    {
+      "seconds": 0.9685989429999609,
+      "bytes": 7522789,
       "pages": 29,
       "shards": 3
     },
     {
-      "seconds": 1.2021151590015506,
-      "bytes": 960345,
-      "pages": 28,
-      "shards": 1
+      "seconds": 0.7894636139972135,
+      "bytes": 5936931,
+      "pages": 37,
+      "shards": 2
     },
     {
-      "seconds": 3.8789173690020107,
-      "bytes": 11941335,
-      "pages": 36,
+      "seconds": 1.0263739810034167,
+      "bytes": 6580240,
+      "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 1.0000528349992237,
-      "bytes": 11258395,
-      "pages": 42,
-      "shards": 4
-    },
-    {
-      "seconds": 0.9514743170002475,
-      "bytes": 822769,
-      "pages": 26,
-      "shards": 1
+      "seconds": 1.228623751005216,
+      "bytes": 6179585,
+      "pages": 37,
+      "shards": 2
     },
     {
-      "seconds": 4.258936992002418,
-      "bytes": 6104340,
-      "pages": 36,
+      "seconds": 1.3004778859976796,
+      "bytes": 12330619,
+      "pages": 35,
       "shards": 3
     },
     {
-      "seconds": 1.0792534860011074,
-      "bytes": 2442489,
-      "pages": 31,
+      "seconds": 1.576678247001837,
+      "bytes": 2288077,
+      "pages": 28,
       "shards": 2
     },
     {
-      "seconds": 1.0146635470009642,
-      "bytes": 11041931,
+      "seconds": 1.2293803360007587,
+      "bytes": 10631332,
       "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 1.0869366869992518,
-      "bytes": 13432151,
-      "pages": 40,
-      "shards": 4
+      "seconds": 0.8141304599994328,
+      "bytes": 2723440,
+      "pages": 38,
+      "shards": 2
     },
     {
-      "seconds": 0.7862899389983795,
-      "bytes": 1239440,
-      "pages": 35,
-      "shards": 1
+      "seconds": 0.9625181489973329,
+      "bytes": 5689945,
+      "pages": 27,
+      "shards": 2
     },
     {
-      "seconds": 0.8729745059972629,
-      "bytes": 5859266,
-      "pages": 29,
+      "seconds": 0.9489649809984257,
+      "bytes": 5862208,
+      "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.0330235869987519,
-      "bytes": 6506267,
-      "pages": 40,
-      "shards": 3
+      "seconds": 1.9020824750041356,
+      "bytes": 12748087,
+      "pages": 37,
+      "shards": 4
     },
     {
-      "seconds": 1.471190406999085,
-      "bytes": 6116938,
-      "pages": 27,
+      "seconds": 0.9002470639970852,
+      "bytes": 6110509,
+      "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 1.1064680679992307,
-      "bytes": 7358899,
-      "pages": 34,
-      "shards": 3
+      "seconds": 0.9653321419973508,
+      "bytes": 880569,
+      "pages": 26,
+      "shards": 1
     },
     {
-      "seconds": 1.2355712200005655,
-      "bytes": 6983067,
-      "pages": 36,
-      "shards": 2
+      "seconds": 1.074353076000989,
+      "bytes": 7774309,
+      "pages": 39,
+      "shards": 3
     },
     {
-      "seconds": 1.2635136329990928,
-      "bytes": 9856957,
-      "pages": 29,
+      "seconds": 1.4574867379997158,
+      "bytes": 9860401,
+      "pages": 33,
       "shards": 3
     },
     {
-      "seconds": 0.961533926998527,
-      "bytes": 492258,
-      "pages": 16,
-      "shards": 1
+      "seconds": 0.7756101689956267,
+      "bytes": 2491834,
+      "pages": 32,
+      "shards": 2
     },
     {
-      "seconds": 1.2022184809975442,
-      "bytes": 7326926,
-      "pages": 31,
+      "seconds": 1.4068310860020574,
+      "bytes": 11759986,
+      "pages": 28,
       "shards": 3
     },
     {
-      "seconds": 0.7280326340005558,
-      "bytes": 1168675,
-      "pages": 35,
+      "seconds": 0.806070794002153,
+      "bytes": 884841,
+      "pages": 27,
       "shards": 1
     },
     {
-      "seconds": 1.1539005899976473,
-      "bytes": 5543356,
-      "pages": 19,
-      "shards": 2
-    },
-    {
-      "seconds": 0.995166409000376,
-      "bytes": 7263909,
+      "seconds": 1.161329097005364,
+      "bytes": 12346007,
       "pages": 39,
-      "shards": 3
+      "shards": 4
     },
     {
-      "seconds": 1.1710582100022293,
-      "bytes": 11639811,
-      "pages": 30,
-      "shards": 3
+      "seconds": 0.916572713998903,
+      "bytes": 942236,
+      "pages": 27,
+      "shards": 1
     },
     {
-      "seconds": 1.5414868839980045,
-      "bytes": 2598188,
+      "seconds": 1.3328421769983834,
+      "bytes": 6157959,
       "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.4500177529989742,
-      "bytes": 6186157,
-      "pages": 35,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1433413880004082,
-      "bytes": 10345973,
-      "pages": 35,
+      "seconds": 1.2523999580007512,
+      "bytes": 7610947,
+      "pages": 30,
       "shards": 3
     },
     {
-      "seconds": 0.963193997999042,
-      "bytes": 2354048,
+      "seconds": 0.7598877909986186,
+      "bytes": 961682,
       "pages": 29,
+      "shards": 1
+    },
+    {
+      "seconds": 1.3535268059931695,
+      "bytes": 16112402,
+      "pages": 42,
+      "shards": 4
+    },
+    {
+      "seconds": 0.9407444530006615,
+      "bytes": 2455446,
+      "pages": 32,
       "shards": 2
     },
     {
-      "seconds": 0.9612779050003155,
-      "bytes": 6227883,
+      "seconds": 0.9131047679984476,
+      "bytes": 6268591,
       "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 0.9593730830019922,
-      "bytes": 6813129,
-      "pages": 31,
-      "shards": 2
+      "seconds": 0.9507441360037774,
+      "bytes": 6652309,
+      "pages": 29,
+      "shards": 3
     },
     {
-      "seconds": 1.1213739670020004,
-      "bytes": 11561927,
-      "pages": 35,
-      "shards": 4
+      "seconds": 0.870253834000323,
+      "bytes": 6235033,
+      "pages": 38,
+      "shards": 3
     },
     {
-      "seconds": 1.0164505040011136,
-      "bytes": 6233089,
-      "pages": 36,
+      "seconds": 0.9296957249971456,
+      "bytes": 6084778,
+      "pages": 29,
       "shards": 2
     },
     {
-      "seconds": 1.140030457998364,
-      "bytes": 2525214,
-      "pages": 34,
-      "shards": 2
+      "seconds": 1.147810253998614,
+      "bytes": 8392070,
+      "pages": 39,
+      "shards": 3
     },
     {
-      "seconds": 0.8727539430001343,
-      "bytes": 5779966,
-      "pages": 32,
-      "shards": 2
+      "seconds": 1.153522004999104,
+      "bytes": 10226084,
+      "pages": 27,
+      "shards": 3
     },
     {
-      "seconds": 1.2494100709991471,
-      "bytes": 10781388,
-      "pages": 37,
+      "seconds": 1.0265658620046452,
+      "bytes": 7737951,
+      "pages": 42,
       "shards": 3
     },
     {
-      "seconds": 0.9040151730005164,
-      "bytes": 8396980,
+      "seconds": 0.7752080549980747,
+      "bytes": 1141313,
       "pages": 35,
-      "shards": 3
+      "shards": 1
     },
     {
-      "seconds": 1.1415953719988465,
-      "bytes": 5808415,
-      "pages": 28,
+      "seconds": 0.8455009660028736,
+      "bytes": 5959516,
+      "pages": 29,
       "shards": 2
     },
     {
-      "seconds": 1.0399577659991337,
-      "bytes": 7078524,
-      "pages": 40,
+      "seconds": 1.0879218709960696,
+      "bytes": 6248481,
+      "pages": 29,
       "shards": 3
     },
     {
-      "seconds": 0.9194039719986904,
-      "bytes": 5850558,
-      "pages": 28,
-      "shards": 2
-    },
-    {
-      "seconds": 0.8959461580016068,
-      "bytes": 6085583,
-      "pages": 36,
+      "seconds": 0.9940221099968767,
+      "bytes": 11924691,
+      "pages": 42,
       "shards": 3
     },
     {
-      "seconds": 0.9613766149996081,
-      "bytes": 2539274,
-      "pages": 33,
+      "seconds": 0.9983105119972606,
+      "bytes": 6041328,
+      "pages": 29,
       "shards": 2
     },
     {
-      "seconds": 1.3802653160018963,
-      "bytes": 12066636,
-      "pages": 34,
-      "shards": 3
+      "seconds": 0.9112700020050397,
+      "bytes": 2626384,
+      "pages": 37,
+      "shards": 2
     },
     {
-      "seconds": 0.828995023999596,
-      "bytes": 4743589,
-      "pages": 35,
-      "shards": 2
+      "seconds": 1.077101428003516,
+      "bytes": 10954533,
+      "pages": 32,
+      "shards": 3
     },
     {
-      "seconds": 0.9685989429999609,
-      "bytes": 7522789,
-      "pages": 29,
+      "seconds": 0.987721411001985,
+      "bytes": 7423511,
+      "pages": 39,
       "shards": 3
     },
     {
-      "seconds": 0.7894636139972135,
-      "bytes": 5936931,
-      "pages": 37,
-      "shards": 2
+      "seconds": 0.7559060139974463,
+      "bytes": 900379,
+      "pages": 27,
+      "shards": 1
     },
     {
-      "seconds": 1.0263739810034167,
-      "bytes": 6580240,
-      "pages": 34,
+      "seconds": 1.208673165005166,
+      "bytes": 11672731,
+      "pages": 32,
       "shards": 3
     },
     {
-      "seconds": 1.228623751005216,
-      "bytes": 6179585,
+      "seconds": 0.8157727979996707,
+      "bytes": 2711312,
       "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.3004778859976796,
-      "bytes": 12330619,
-      "pages": 35,
+      "seconds": 1.0441342500052997,
+      "bytes": 9648084,
+      "pages": 31,
       "shards": 3
     },
     {
-      "seconds": 1.576678247001837,
-      "bytes": 2288077,
-      "pages": 28,
+      "seconds": 0.8669746300001862,
+      "bytes": 6281304,
+      "pages": 38,
       "shards": 2
     },
     {
-      "seconds": 1.2293803360007587,
-      "bytes": 10631332,
-      "pages": 34,
+      "seconds": 1.0493857530018431,
+      "bytes": 2405359,
+      "pages": 29,
+      "shards": 2
+    },
+    {
+      "seconds": 1.4177541469980497,
+      "bytes": 10495082,
+      "pages": 41,
       "shards": 3
     },
     {
-      "seconds": 0.8141304599994328,
-      "bytes": 2723440,
-      "pages": 38,
-      "shards": 2
+      "seconds": 0.93050350699923,
+      "bytes": 8304633,
+      "pages": 32,
+      "shards": 3
     },
     {
-      "seconds": 0.9625181489973329,
-      "bytes": 5689945,
-      "pages": 27,
+      "seconds": 1.291375706998224,
+      "bytes": 5813086,
+      "pages": 29,
       "shards": 2
     },
     {
-      "seconds": 0.9489649809984257,
-      "bytes": 5862208,
-      "pages": 37,
+      "seconds": 1.007629627994902,
+      "bytes": 6351245,
+      "pages": 35,
       "shards": 2
     },
     {
-      "seconds": 1.9020824750041356,
-      "bytes": 12748087,
-      "pages": 37,
-      "shards": 4
+      "seconds": 1.1004838699955144,
+      "bytes": 7253121,
+      "pages": 34,
+      "shards": 3
     },
     {
-      "seconds": 0.9002470639970852,
-      "bytes": 6110509,
-      "pages": 36,
+      "seconds": 0.860785910001141,
+      "bytes": 6150922,
+      "pages": 38,
       "shards": 2
     },
     {
-      "seconds": 0.9653321419973508,
-      "bytes": 880569,
-      "pages": 26,
-      "shards": 1
+      "seconds": 0.967503943000338,
+      "bytes": 2353719,
+      "pages": 29,
+      "shards": 2
     },
     {
-      "seconds": 1.074353076000989,
-      "bytes": 7774309,
-      "pages": 39,
+      "seconds": 1.1848933270011912,
+      "bytes": 11480176,
+      "pages": 30,
       "shards": 3
     },
     {
-      "seconds": 1.4574867379997158,
-      "bytes": 9860401,
-      "pages": 33,
+      "seconds": 0.9306560840050224,
+      "bytes": 5914010,
+      "pages": 31,
+      "shards": 2
+    },
+    {
+      "seconds": 0.986767267997493,
+      "bytes": 6142895,
+      "pages": 38,
       "shards": 3
     }
   ],

trainer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5026bd4bb1b3b18f5fc726c01908ba56660bc127713bb70b791880fb27548eeb
-size 721791615
+oid sha256:63c9c8c99731fc20714a54a44377517fe0f1f3043beea6e7f6b8a0c370badb1e
+size 721775039

training_config.json

@@ -46,8 +46,8 @@
   "cpu_threads": 4,
   "page_rows": 4096,
   "cortex_page_rows": 128,
-  "prefetch_batches": 61,
-  "prefetch_low_water": 30,
+  "prefetch_batches": 62,
+  "prefetch_low_water": 31,
   "memory_pages": 12,
   "cache_gb": 20,
   "force_rebenchmark": false,