Qwen90M step 670000; complete resumable checkpoint

complete.json

@@ -1,18 +1,18 @@
 {
-  "step": 665000,
+  "step": 670000,
   "files": {
     "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
     "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
-    "model.safetensors": "6c241dc1c1a846d32a12afafc492600d69a86e3be4fee8d141a618e8afd44783",
+    "model.safetensors": "3e61d5e203d0bc4da303d0c73f37e2c56ee4a56703224613ed6b408956f545c6",
     "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
     "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
     "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
     "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
-    "trainer.pt": "717cb301aeb14d38409ae0e0d04856c0c71f8067a7c0afbf8c05d0053df1f880",
-    "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb",
+    "trainer.pt": "baf181694b42eed96c2f274abdcd70b231be691131559b6fae917122d444a98d",
+    "training_config.json": "99d98cec9c0814e63e7b7a5836277d5c336d3c2fb33fa69da403f3610c902c99",
     "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf",
-    "shard_scheduler.json": "e7d4e34ca9c5186534fb32e800aad10548f8a5c95b53fbc5af10a7ccc7febab0",
-    "coverage.json": "ed08e0edb4aa125a6baee640d15e6403fa05eecc9fd2bba36beb390a00d0f15e",
+    "shard_scheduler.json": "7d1f5ece6f823a90937aac50d59b702f0bdefd7399ebea8da1b48bbc8a86ccdf",
+    "coverage.json": "7cfc71bc0efed1fc9ac9186a78aa9beee721c934d777b8a3ea87a912a1cbb4df",
     "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
   },
   "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a"

coverage.json

@@ -1,107 +1,107 @@
 {
-  "step": 665000,
+  "step": 670000,
   "stats": {
-    "climbmix/blocks": 23940000,
-    "climbmix/targets": 40021124917,
-    "climbmix/input_tokens": 40023660910,
-    "ultra_qa/blocks": 8977500,
-    "ultra_qa/targets": 14646361182,
-    "ultra_qa/input_tokens": 14646389468,
-    "cortex/blocks": 8977500,
-    "cortex/targets": 341355085,
-    "cortex/input_tokens": 14270346565,
-    "ultra_style/blocks": 8977500,
-    "ultra_style/targets": 14200262565,
-    "ultra_style/input_tokens": 14201007196,
-    "openmath/blocks": 5985000,
-    "openmath/targets": 9247885163,
-    "openmath/input_tokens": 9250966078,
-    "rewrite6/blocks": 2992500,
-    "rewrite6/targets": 4905783456,
-    "rewrite6/input_tokens": 4911161965
+    "climbmix/blocks": 24120000,
+    "climbmix/targets": 40322063590,
+    "climbmix/input_tokens": 40324618718,
+    "ultra_qa/blocks": 9045000,
+    "ultra_qa/targets": 14756386919,
+    "ultra_qa/input_tokens": 14756415423,
+    "cortex/blocks": 9045000,
+    "cortex/targets": 343924602,
+    "cortex/input_tokens": 14377667142,
+    "ultra_style/blocks": 9045000,
+    "ultra_style/targets": 14306999204,
+    "ultra_style/input_tokens": 14307749781,
+    "openmath/blocks": 6030000,
+    "openmath/targets": 9317442826,
+    "openmath/input_tokens": 9320546731,
+    "rewrite6/blocks": 3015000,
+    "rewrite6/targets": 4942621070,
+    "rewrite6/input_tokens": 4948040293
   },
   "sources": {
     "climbmix": {
-      "page": 3278,
-      "row": 1762,
+      "page": 3303,
+      "row": 479,
       "start": {
-        "file": 158,
+        "file": 159,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 55,
+          "row_group": 73,
           "row": 1024
         }
       },
       "epoch": 0,
-      "offset": 0,
-      "documents": 13428450
+      "offset": 14336,
+      "documents": 13529567
     },
     "rewrite6": {
-      "page": 15,
-      "row": 1069,
+      "page": 29,
+      "row": 1718,
       "start": {
         "file": 0,
         "reader": {
           "format": "jsonl-row-v1",
-          "row": 61440
+          "row": 118784
         }
       },
       "epoch": 37,
-      "offset": 2048,
-      "documents": 7662457
+      "offset": 0,
+      "documents": 7720450
     },
     "openmath": {
-      "page": 2048,
-      "row": 3984,
+      "page": 2064,
+      "row": 1413,
       "start": {
         "file": 19,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 92,
-          "row": 258
+          "row_group": 157,
+          "row": 794
         }
       },
       "epoch": 0,
       "offset": 0,
-      "documents": 8392592
+      "documents": 8455557
     },
     "ultra_qa": {
-      "page": 886,
-      "row": 34,
+      "page": 892,
+      "row": 2846,
       "start": {
-        "file": 6,
+        "file": 7,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 7,
-          "row": 59275
+          "row_group": 0,
+          "row": 16248
         }
       },
       "epoch": 0,
       "offset": 2048,
-      "documents": 3629090
+      "documents": 3656478
     },
     "ultra_style": {
-      "page": 1352,
-      "row": 176,
+      "page": 1362,
+      "row": 1579,
       "start": {
         "file": 8,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 0,
-          "row": 58258
+          "row_group": 1,
+          "row": 14660
         }
       },
       "epoch": 0,
       "offset": 2048,
-      "documents": 5537968
+      "documents": 5580331
     },
     "cortex": {
-      "page": 174111,
-      "row": 108,
+      "page": 175421,
+      "row": 25,
       "start": null,
       "epoch": 0,
       "offset": 0,
-      "documents": 22286316
+      "documents": 22453913
     }
   }
 }

latest.json

@@ -1 +1 @@
-{"step": 665000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "6c241dc1c1a846d32a12afafc492600d69a86e3be4fee8d141a618e8afd44783", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "717cb301aeb14d38409ae0e0d04856c0c71f8067a7c0afbf8c05d0053df1f880", "training_config.json": "6c3c3ccff89dabb5098a22bc207c2f3abaee8b3bffec3f6cf6ad361a63b1a7fb", "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf", "shard_scheduler.json": "e7d4e34ca9c5186534fb32e800aad10548f8a5c95b53fbc5af10a7ccc7febab0", "coverage.json": "ed08e0edb4aa125a6baee640d15e6403fa05eecc9fd2bba36beb390a00d0f15e", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000665000"}
+{"step": 670000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "3e61d5e203d0bc4da303d0c73f37e2c56ee4a56703224613ed6b408956f545c6", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "baf181694b42eed96c2f274abdcd70b231be691131559b6fae917122d444a98d", "training_config.json": "99d98cec9c0814e63e7b7a5836277d5c336d3c2fb33fa69da403f3610c902c99", "performance.json": "5949473856ba60c6ea6e09958b880f6c16bf1db77e55259c9de9f88123633edf", "shard_scheduler.json": "7d1f5ece6f823a90937aac50d59b702f0bdefd7399ebea8da1b48bbc8a86ccdf", "coverage.json": "7cfc71bc0efed1fc9ac9186a78aa9beee721c934d777b8a3ea87a912a1cbb4df", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "33a459803c8cffa8929a90def998b4ad2c7c52fd1245365a9ee946b5552d845a", "folder": "step-000000670000"}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6c241dc1c1a846d32a12afafc492600d69a86e3be4fee8d141a618e8afd44783
+oid sha256:3e61d5e203d0bc4da303d0c73f37e2c56ee4a56703224613ed6b408956f545c6
 size 360622744

shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77402.7761039638,
-      "required_pages_per_second_upper_estimate": 0.005354903650056715
+      "required_token_positions_per_second": 74846.34535141522,
+      "required_pages_per_second_upper_estimate": 0.0051780438388588255
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9675.347012995475,
-      "required_pages_per_second_upper_estimate": 0.00405086575012036
+      "required_token_positions_per_second": 9355.793168926903,
+      "required_pages_per_second_upper_estimate": 0.00391707522865193
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19350.69402599095,
-      "required_pages_per_second_upper_estimate": 0.004030239253495446
+      "required_token_positions_per_second": 18711.586337853805,
+      "required_pages_per_second_upper_estimate": 0.003897129976459624
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29026.041038986423,
-      "required_pages_per_second_upper_estimate": 0.0017400668400887923
+      "required_token_positions_per_second": 28067.379506780704,
+      "required_pages_per_second_upper_estimate": 0.0016825965450244628
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29026.041038986423,
-      "required_pages_per_second_upper_estimate": 0.002256647018719463
+      "required_token_positions_per_second": 28067.379506780704,
+      "required_pages_per_second_upper_estimate": 0.0021821153012969137
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29026.041038986423,
-      "required_pages_per_second_upper_estimate": 0.3723148887134134
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28067.379506780704,
+      "required_pages_per_second_upper_estimate": 0.36001820790883526
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,413 +155,413 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08651116098442913,
-    "producer_p95_seconds": 0.24094921500000055,
-    "gpu_update_seconds": 0.9525239753800561,
-    "producer_batches_per_second": 11.559202172538026,
-    "gpu_batches_per_second": 1.0498423408198214,
-    "producer_headroom": 11.0104172056077,
+    "producer_batch_seconds": 0.10544663335164728,
+    "producer_p95_seconds": 0.27246575100070913,
+    "gpu_update_seconds": 0.9850581167835997,
+    "producer_batches_per_second": 9.483470151817588,
+    "gpu_batches_per_second": 1.0151685296144641,
+    "producer_headroom": 9.341769248322912,
     "sustainable": true,
-    "buffer_seconds": 60.009010448943535,
+    "buffer_seconds": 60.08854512379958,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "shard_target_mib": 32,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357946.5645164134,
-    "upload_bytes_per_second": 1903783.0719585211,
+    "generation_bytes_per_second": 666047.1250430234,
+    "upload_bytes_per_second": 5380735.178534467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.80700017566014,
+    "estimated_full_shards_per_hour": 71.45910412534727,
     "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": 1.164695950999885,
-      "bytes": 11623258,
+      "seconds": 0.9111081320006633,
+      "bytes": 972942,
+      "pages": 31,
+      "shards": 1
+    },
+    {
+      "seconds": 1.5192261520023749,
+      "bytes": 17166243,
       "pages": 34,
-      "shards": 3
+      "shards": 5
     },
     {
-      "seconds": 1.1422205050002958,
-      "bytes": 2673736,
-      "pages": 36,
-      "shards": 3
+      "seconds": 0.9617387159996724,
+      "bytes": 1097341,
+      "pages": 34,
+      "shards": 1
     },
     {
-      "seconds": 1.4229377800002112,
-      "bytes": 5867572,
-      "pages": 28,
+      "seconds": 1.1497930719997385,
+      "bytes": 6229541,
+      "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 1.0479602760005946,
-      "bytes": 10355941,
-      "pages": 43,
+      "seconds": 1.2241392690011708,
+      "bytes": 7522866,
+      "pages": 31,
       "shards": 3
     },
     {
-      "seconds": 1.0539684499999566,
-      "bytes": 2291279,
-      "pages": 28,
+      "seconds": 1.1316358820004098,
+      "bytes": 5807643,
+      "pages": 41,
       "shards": 2
     },
     {
-      "seconds": 1.055462077001721,
-      "bytes": 6384191,
-      "pages": 37,
+      "seconds": 1.0283135649988253,
+      "bytes": 5734251,
+      "pages": 28,
       "shards": 2
     },
     {
-      "seconds": 1.3574712270001328,
-      "bytes": 12582265,
-      "pages": 37,
+      "seconds": 1.5221587859996362,
+      "bytes": 13580138,
+      "pages": 41,
       "shards": 4
     },
     {
-      "seconds": 1.1109151510008815,
-      "bytes": 6070099,
-      "pages": 37,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0409215150011732,
-      "bytes": 6007655,
+      "seconds": 1.401579060999211,
+      "bytes": 925271,
       "pages": 28,
-      "shards": 2
+      "shards": 1
     },
     {
-      "seconds": 0.9865703520008537,
-      "bytes": 2607815,
+      "seconds": 1.7116284690018801,
+      "bytes": 11540303,
       "pages": 37,
-      "shards": 2
+      "shards": 4
     },
     {
-      "seconds": 1.1019334159991558,
-      "bytes": 10289866,
-      "pages": 39,
-      "shards": 3
+      "seconds": 0.9012360309970973,
+      "bytes": 1247702,
+      "pages": 35,
+      "shards": 1
     },
     {
-      "seconds": 1.2059528549998504,
-      "bytes": 7149120,
-      "pages": 30,
+      "seconds": 1.5333594760013511,
+      "bytes": 7425995,
+      "pages": 38,
       "shards": 3
     },
     {
-      "seconds": 0.9931068189998769,
-      "bytes": 6968332,
-      "pages": 33,
-      "shards": 2
-    },
-    {
-      "seconds": 1.032158955000341,
-      "bytes": 6117925,
-      "pages": 37,
+      "seconds": 1.141444616998342,
+      "bytes": 5373163,
+      "pages": 23,
       "shards": 2
     },
     {
-      "seconds": 122.56198037699869,
-      "bytes": 6934679,
-      "pages": 34,
+      "seconds": 1.302073504000873,
+      "bytes": 12612142,
+      "pages": 45,
       "shards": 3
     },
     {
-      "seconds": 1.1427723569995578,
-      "bytes": 13775084,
-      "pages": 72,
+      "seconds": 1.2979291070005274,
+      "bytes": 7799504,
+      "pages": 35,
       "shards": 3
     },
     {
-      "seconds": 1.2821505469983094,
-      "bytes": 1156096,
-      "pages": 35,
+      "seconds": 0.9238773860015499,
+      "bytes": 1050494,
+      "pages": 29,
       "shards": 1
     },
     {
-      "seconds": 1.3412859470008698,
-      "bytes": 15865632,
+      "seconds": 1.0488793139993504,
+      "bytes": 6030424,
       "pages": 31,
-      "shards": 4
+      "shards": 2
     },
     {
-      "seconds": 1.324599482000849,
-      "bytes": 7784072,
-      "pages": 39,
-      "shards": 3
+      "seconds": 1.2485899509993033,
+      "bytes": 11733880,
+      "pages": 45,
+      "shards": 4
     },
     {
-      "seconds": 0.8532274739991408,
-      "bytes": 1099982,
-      "pages": 34,
+      "seconds": 1.0325014549998741,
+      "bytes": 941600,
+      "pages": 26,
       "shards": 1
     },
     {
-      "seconds": 1.0939723810006399,
-      "bytes": 5792220,
-      "pages": 28,
-      "shards": 2
+      "seconds": 1.327213976001076,
+      "bytes": 13471794,
+      "pages": 42,
+      "shards": 4
     },
     {
-      "seconds": 1.519392225000047,
-      "bytes": 12323148,
-      "pages": 43,
-      "shards": 4
+      "seconds": 1.1652102200023364,
+      "bytes": 5705697,
+      "pages": 29,
+      "shards": 2
     },
     {
-      "seconds": 0.8898182180000731,
-      "bytes": 896140,
-      "pages": 26,
-      "shards": 1
+      "seconds": 1.1611206180023146,
+      "bytes": 6127791,
+      "pages": 36,
+      "shards": 3
     },
     {
-      "seconds": 1.8745819040013885,
-      "bytes": 13367716,
-      "pages": 39,
-      "shards": 4
+      "seconds": 1.3078043249988696,
+      "bytes": 7396085,
+      "pages": 30,
+      "shards": 3
     },
     {
-      "seconds": 0.9403650779986492,
-      "bytes": 1055784,
-      "pages": 32,
+      "seconds": 0.9337214369988942,
+      "bytes": 1171128,
+      "pages": 35,
       "shards": 1
     },
     {
-      "seconds": 1.317758003999188,
-      "bytes": 10010526,
-      "pages": 18,
-      "shards": 3
+      "seconds": 1.1392940650002856,
+      "bytes": 6258107,
+      "pages": 31,
+      "shards": 2
     },
     {
-      "seconds": 1.4487697540007503,
-      "bytes": 2537038,
-      "pages": 36,
-      "shards": 2
+      "seconds": 1.18578564299969,
+      "bytes": 11925039,
+      "pages": 24,
+      "shards": 4
     },
     {
-      "seconds": 1.5384773589994438,
-      "bytes": 6037444,
-      "pages": 29,
+      "seconds": 1.0414635330016608,
+      "bytes": 5963906,
+      "pages": 28,
       "shards": 2
     },
     {
-      "seconds": 0.9018981910012371,
-      "bytes": 1284941,
+      "seconds": 1.0295740060028038,
+      "bytes": 1098382,
       "pages": 35,
       "shards": 1
     },
     {
-      "seconds": 1.8045908780004538,
-      "bytes": 17039792,
+      "seconds": 1.3453053759985778,
+      "bytes": 7563007,
       "pages": 37,
-      "shards": 5
+      "shards": 3
     },
     {
-      "seconds": 1.6887772259997291,
-      "bytes": 6318440,
+      "seconds": 1.2517445409976062,
+      "bytes": 10940783,
+      "pages": 32,
+      "shards": 3
+    },
+    {
+      "seconds": 1.2907915830001002,
+      "bytes": 2619734,
       "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.1342329689996404,
-      "bytes": 2581964,
-      "pages": 37,
+      "seconds": 1.1900975380012824,
+      "bytes": 11578816,
+      "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 1.048425811000925,
-      "bytes": 5676510,
-      "pages": 27,
-      "shards": 2
+      "seconds": 1.276093910997588,
+      "bytes": 903821,
+      "pages": 26,
+      "shards": 1
     },
     {
-      "seconds": 33.502104078999764,
-      "bytes": 11160724,
-      "pages": 35,
-      "shards": 3
+      "seconds": 1.6412474520002434,
+      "bytes": 11839721,
+      "pages": 42,
+      "shards": 4
     },
     {
-      "seconds": 0.9322519389988884,
-      "bytes": 2962222,
-      "pages": 46,
+      "seconds": 1.4279512529974454,
+      "bytes": 5596025,
+      "pages": 31,
       "shards": 2
     },
     {
-      "seconds": 1.3943822889996227,
-      "bytes": 11929043,
-      "pages": 39,
-      "shards": 3
+      "seconds": 1.1274023559999478,
+      "bytes": 1023034,
+      "pages": 31,
+      "shards": 1
     },
     {
-      "seconds": 1.140193093000562,
-      "bytes": 7462921,
-      "pages": 30,
+      "seconds": 1.0909878259990364,
+      "bytes": 7709457,
+      "pages": 39,
       "shards": 3
     },
     {
-      "seconds": 1.6433775559999049,
-      "bytes": 5520234,
-      "pages": 33,
-      "shards": 2
+      "seconds": 2.9601359040025272,
+      "bytes": 15395500,
+      "pages": 35,
+      "shards": 4
     },
     {
-      "seconds": 1.1214206500008004,
-      "bytes": 6053975,
+      "seconds": 1.3234173230011947,
+      "bytes": 2631414,
       "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.0533692200006044,
-      "bytes": 7678729,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 0.9111081320006633,
-      "bytes": 972942,
-      "pages": 31,
-      "shards": 1
+      "seconds": 1.286528674998408,
+      "bytes": 6199355,
+      "pages": 28,
+      "shards": 2
     },
     {
-      "seconds": 1.5192261520023749,
-      "bytes": 17166243,
-      "pages": 34,
-      "shards": 5
+      "seconds": 1.3863541229984548,
+      "bytes": 10474994,
+      "pages": 40,
+      "shards": 3
     },
     {
-      "seconds": 0.9617387159996724,
-      "bytes": 1097341,
-      "pages": 34,
-      "shards": 1
+      "seconds": 1.2548714610020397,
+      "bytes": 2378266,
+      "pages": 29,
+      "shards": 2
     },
     {
-      "seconds": 1.1497930719997385,
-      "bytes": 6229541,
-      "pages": 36,
+      "seconds": 1.0437052470006165,
+      "bytes": 6102982,
+      "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.2241392690011708,
-      "bytes": 7522866,
-      "pages": 31,
+      "seconds": 1.201682383001753,
+      "bytes": 8258960,
+      "pages": 32,
       "shards": 3
     },
     {
-      "seconds": 1.1316358820004098,
-      "bytes": 5807643,
-      "pages": 41,
-      "shards": 2
+      "seconds": 1.4597207000006165,
+      "bytes": 10914765,
+      "pages": 43,
+      "shards": 3
     },
     {
-      "seconds": 1.0283135649988253,
-      "bytes": 5734251,
+      "seconds": 1.293525116998353,
+      "bytes": 5848497,
       "pages": 28,
       "shards": 2
     },
     {
-      "seconds": 1.5221587859996362,
-      "bytes": 13580138,
-      "pages": 41,
-      "shards": 4
-    },
-    {
-      "seconds": 1.401579060999211,
-      "bytes": 925271,
-      "pages": 28,
-      "shards": 1
+      "seconds": 1.0607272609995562,
+      "bytes": 2667610,
+      "pages": 38,
+      "shards": 2
     },
     {
-      "seconds": 1.7116284690018801,
-      "bytes": 11540303,
-      "pages": 37,
-      "shards": 4
+      "seconds": 1.5769093039998552,
+      "bytes": 5962126,
+      "pages": 29,
+      "shards": 2
     },
     {
-      "seconds": 0.9012360309970973,
-      "bytes": 1247702,
-      "pages": 35,
-      "shards": 1
+      "seconds": 1.0608517170003324,
+      "bytes": 4607080,
+      "pages": 31,
+      "shards": 2
     },
     {
-      "seconds": 1.5333594760013511,
-      "bytes": 7425995,
-      "pages": 38,
-      "shards": 3
+      "seconds": 1.4511087869977928,
+      "bytes": 13183448,
+      "pages": 39,
+      "shards": 4
     },
     {
-      "seconds": 1.141444616998342,
-      "bytes": 5373163,
-      "pages": 23,
+      "seconds": 1.3023088439986168,
+      "bytes": 6313993,
+      "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 1.302073504000873,
-      "bytes": 12612142,
-      "pages": 45,
-      "shards": 3
+      "seconds": 1.0238163379981415,
+      "bytes": 2362328,
+      "pages": 29,
+      "shards": 2
     },
     {
-      "seconds": 1.2979291070005274,
-      "bytes": 7799504,
-      "pages": 35,
-      "shards": 3
+      "seconds": 1.3659524169997894,
+      "bytes": 10315345,
+      "pages": 40,
+      "shards": 4
     },
     {
-      "seconds": 0.9238773860015499,
-      "bytes": 1050494,
-      "pages": 29,
-      "shards": 1
+      "seconds": 1.0596161510002275,
+      "bytes": 5840926,
+      "pages": 28,
+      "shards": 2
     },
     {
-      "seconds": 1.0488793139993504,
-      "bytes": 6030424,
-      "pages": 31,
+      "seconds": 0.9319052459977684,
+      "bytes": 2836860,
+      "pages": 38,
       "shards": 2
     },
     {
-      "seconds": 1.2485899509993033,
-      "bytes": 11733880,
-      "pages": 45,
+      "seconds": 1.3332925520007848,
+      "bytes": 16563668,
+      "pages": 36,
       "shards": 4
     },
     {
-      "seconds": 1.0325014549998741,
-      "bytes": 941600,
-      "pages": 26,
-      "shards": 1
+      "seconds": 1.482665261002694,
+      "bytes": 2527538,
+      "pages": 33,
+      "shards": 2
     },
     {
-      "seconds": 1.327213976001076,
-      "bytes": 13471794,
-      "pages": 42,
-      "shards": 4
+      "seconds": 1.213385584000207,
+      "bytes": 6122336,
+      "pages": 33,
+      "shards": 2
     },
     {
-      "seconds": 1.1652102200023364,
-      "bytes": 5705697,
-      "pages": 29,
+      "seconds": 1.3804073209976195,
+      "bytes": 6371391,
+      "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.1611206180023146,
-      "bytes": 6127791,
+      "seconds": 1.1654375459984294,
+      "bytes": 2648981,
       "pages": 36,
-      "shards": 3
+      "shards": 2
     },
     {
-      "seconds": 1.3078043249988696,
-      "bytes": 7396085,
-      "pages": 30,
+      "seconds": 1.4223995709980954,
+      "bytes": 9800135,
+      "pages": 33,
       "shards": 3
     },
     {
-      "seconds": 0.9337214369988942,
-      "bytes": 1171128,
+      "seconds": 1.239405350999732,
+      "bytes": 13306132,
       "pages": 35,
+      "shards": 4
+    },
+    {
+      "seconds": 1.4411270890013839,
+      "bytes": 1188235,
+      "pages": 36,
       "shards": 1
+    },
+    {
+      "seconds": 1.200595816000714,
+      "bytes": 10503607,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

trainer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:717cb301aeb14d38409ae0e0d04856c0c71f8067a7c0afbf8c05d0053df1f880
-size 721779967
+oid sha256:baf181694b42eed96c2f274abdcd70b231be691131559b6fae917122d444a98d
+size 721763775

training_config.json

@@ -53,8 +53,8 @@
   "cpu_threads": 4,
   "page_rows": 4096,
   "cortex_page_rows": 128,
-  "prefetch_batches": 63,
-  "prefetch_low_water": 31,
+  "prefetch_batches": 61,
+  "prefetch_low_water": 30,
   "memory_pages": 12,
   "cache_gb": 20,
   "force_rebenchmark": false,