Qwen90M step 705000; complete resumable checkpoint

complete.json

@@ -1,18 +1,18 @@
 {
-  "step": 700000,
+  "step": 705000,
   "files": {
     "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
     "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
-    "model.safetensors": "d2961c6f05c442c4ab734fdc321c4bd99d4d2cdb08dfc0df55b424ae0cb668fd",
+    "model.safetensors": "c24a63cd2e1cb9edf30d6409a48273ab6c01c49122624fe51ba44135f928c526",
     "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
     "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
     "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
     "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
-    "trainer.pt": "071781bcfaf5f9809ccf892899e3b350ca09fb10a20d4e9fa8aca6156c9779ea",
-    "training_config.json": "6b17d8d96cc52df4441b50c79edf3e942bfe52f3c28b64d97f8a592a94b8549e",
+    "trainer.pt": "7d0f9cff5ba25bc9ee4e96e42582ae75644b70f80f581eeb59cc1552ebe7c806",
+    "training_config.json": "78af77f846fb3dcba08002ed692d24a6e4b3900ba35e5592c606aca3466b41f8",
     "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29",
-    "shard_scheduler.json": "cc4225a4f4c10a3ebe7ae762beddc0b1028f2619039eed4fbdfc67099a29a929",
-    "coverage.json": "fbe12c0005986d94f81200adeea5b3161fe8c1a4817c2774a13be90699b44102",
+    "shard_scheduler.json": "92f5af91d86da30a207fc2ede8db0f46e2f61efa3eee22a9c1c2a58df9954b09",
+    "coverage.json": "8a4c9c5871c780362c346a7c8a31c7e57337e0f11d467c16d825aefec83f9598",
     "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
   },
   "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac"

coverage.json

@@ -1,107 +1,107 @@
 {
-  "step": 700000,
+  "step": 705000,
   "stats": {
-    "climbmix/blocks": 25200000,
-    "climbmix/targets": 42126492444,
-    "climbmix/input_tokens": 42129162281,
-    "ultra_qa/blocks": 9450000,
-    "ultra_qa/targets": 15416907813,
-    "ultra_qa/input_tokens": 15416937583,
-    "cortex/blocks": 9450000,
-    "cortex/targets": 359315602,
-    "cortex/input_tokens": 15021391599,
-    "ultra_style/blocks": 9450000,
-    "ultra_style/targets": 14946958997,
-    "ultra_style/input_tokens": 14947743302,
-    "openmath/blocks": 6300000,
-    "openmath/targets": 9734892069,
-    "openmath/input_tokens": 9738135363,
-    "rewrite6/blocks": 3150000,
-    "rewrite6/targets": 5163995804,
-    "rewrite6/input_tokens": 5169657315
+    "climbmix/blocks": 25380000,
+    "climbmix/targets": 42427427006,
+    "climbmix/input_tokens": 42430116101,
+    "ultra_qa/blocks": 9517500,
+    "ultra_qa/targets": 15526937657,
+    "ultra_qa/input_tokens": 15526967636,
+    "cortex/blocks": 9517500,
+    "cortex/targets": 361880143,
+    "cortex/input_tokens": 15128656803,
+    "ultra_style/blocks": 9517500,
+    "ultra_style/targets": 15053668730,
+    "ultra_style/input_tokens": 15054458776,
+    "openmath/blocks": 6345000,
+    "openmath/targets": 9804346129,
+    "openmath/input_tokens": 9807612496,
+    "rewrite6/blocks": 3172500,
+    "rewrite6/targets": 5200843675,
+    "rewrite6/input_tokens": 5206545918
   },
   "sources": {
     "climbmix": {
-      "page": 3451,
-      "row": 1860,
+      "page": 3476,
+      "row": 737,
       "start": {
-        "file": 167,
+        "file": 168,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 2,
+          "row_group": 19,
           "row": 1024
         }
       },
       "epoch": 0,
-      "offset": 2048,
-      "documents": 14137156
+      "offset": 0,
+      "documents": 14238433
     },
     "rewrite6": {
-      "page": 13,
-      "row": 1638,
+      "page": 27,
+      "row": 2327,
       "start": {
         "file": 0,
         "reader": {
           "format": "jsonl-row-v1",
-          "row": 53248
+          "row": 110592
         }
       },
       "epoch": 39,
       "offset": 0,
-      "documents": 8065642
+      "documents": 8123675
     },
     "openmath": {
-      "page": 2156,
-      "row": 3724,
+      "page": 2172,
+      "row": 1176,
       "start": {
         "file": 20,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 97,
-          "row": 976
+          "row_group": 163,
+          "row": 512
         }
       },
       "epoch": 0,
-      "offset": 2048,
-      "documents": 8834700
+      "offset": 0,
+      "documents": 8897688
     },
     "ultra_qa": {
-      "page": 932,
-      "row": 2614,
+      "page": 939,
+      "row": 1127,
       "start": {
         "file": 7,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 2,
-          "row": 50805
+          "row_group": 3,
+          "row": 15078
         }
       },
       "epoch": 0,
       "offset": 2048,
-      "documents": 3820086
+      "documents": 3847271
     },
     "ultra_style": {
-      "page": 1423,
-      "row": 2123,
+      "page": 1433,
+      "row": 3105,
       "start": {
         "file": 8,
         "reader": {
           "format": "parquet-row-v1",
           "row_group": 4,
-          "row": 8311
+          "row": 49271
         }
       },
       "epoch": 0,
       "offset": 2048,
-      "documents": 5830731
+      "documents": 5872673
     },
     "cortex": {
-      "page": 183275,
-      "row": 1,
+      "page": 184583,
+      "row": 79,
       "start": null,
       "epoch": 0,
       "offset": 0,
-      "documents": 23459201
+      "documents": 23626703
     }
   }
 }

latest.json

@@ -1 +1 @@
-{"step": 700000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "d2961c6f05c442c4ab734fdc321c4bd99d4d2cdb08dfc0df55b424ae0cb668fd", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "071781bcfaf5f9809ccf892899e3b350ca09fb10a20d4e9fa8aca6156c9779ea", "training_config.json": "6b17d8d96cc52df4441b50c79edf3e942bfe52f3c28b64d97f8a592a94b8549e", "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29", "shard_scheduler.json": "cc4225a4f4c10a3ebe7ae762beddc0b1028f2619039eed4fbdfc67099a29a929", "coverage.json": "fbe12c0005986d94f81200adeea5b3161fe8c1a4817c2774a13be90699b44102", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac", "folder": "step-000000700000"}
+{"step": 705000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "c24a63cd2e1cb9edf30d6409a48273ab6c01c49122624fe51ba44135f928c526", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "7d0f9cff5ba25bc9ee4e96e42582ae75644b70f80f581eeb59cc1552ebe7c806", "training_config.json": "78af77f846fb3dcba08002ed692d24a6e4b3900ba35e5592c606aca3466b41f8", "performance.json": "96c7d3993f53da33aeba321717c341456e5878e35efa676b6ffe17d014497e29", "shard_scheduler.json": "92f5af91d86da30a207fc2ede8db0f46e2f61efa3eee22a9c1c2a58df9954b09", "coverage.json": "8a4c9c5871c780362c346a7c8a31c7e57337e0f11d467c16d825aefec83f9598", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "1dee8333ad826e9fb3d197c33535515cdb9de560b94f624cae15e044b464f0ac", "folder": "step-000000705000"}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d2961c6f05c442c4ab734fdc321c4bd99d4d2cdb08dfc0df55b424ae0cb668fd
+oid sha256:c24a63cd2e1cb9edf30d6409a48273ab6c01c49122624fe51ba44135f928c526
 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": 77507.70753695644,
-      "required_pages_per_second_upper_estimate": 0.005362163050065615
+      "required_token_positions_per_second": 77341.28166839482,
+      "required_pages_per_second_upper_estimate": 0.005350649322317305
     },
     {
       "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": 9688.463442119555,
-      "required_pages_per_second_upper_estimate": 0.004056357325092425
+      "required_token_positions_per_second": 9667.660208549352,
+      "required_pages_per_second_upper_estimate": 0.004047647445617498
     },
     {
       "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": 19376.92688423911,
-      "required_pages_per_second_upper_estimate": 0.004035702866061544
+      "required_token_positions_per_second": 19335.320417098705,
+      "required_pages_per_second_upper_estimate": 0.0040270373361925215
     },
     {
       "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": 29065.39032635867,
-      "required_pages_per_second_upper_estimate": 0.0017424257697838755
+      "required_token_positions_per_second": 29002.98062564806,
+      "required_pages_per_second_upper_estimate": 0.0017386844035203698
     },
     {
       "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": 29065.39032635867,
-      "required_pages_per_second_upper_estimate": 0.002259706252733432
+      "required_token_positions_per_second": 29002.98062564806,
+      "required_pages_per_second_upper_estimate": 0.002254854173014444
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 29065.39032635867,
-      "required_pages_per_second_upper_estimate": 0.3728196191218516
+      "required_token_positions_per_second": 29002.98062564806,
+      "required_pages_per_second_upper_estimate": 0.3720190944914516
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08243451003123425,
-    "producer_p95_seconds": 0.21277156500036654,
-    "gpu_update_seconds": 0.9512344300061482,
-    "producer_batches_per_second": 12.130841799400544,
-    "gpu_batches_per_second": 1.0512655644660978,
-    "producer_headroom": 11.539274384547534,
+    "producer_batch_seconds": 0.09527713949219674,
+    "producer_p95_seconds": 0.2191641060016991,
+    "gpu_update_seconds": 0.9532813319038729,
+    "producer_batches_per_second": 10.495697135007928,
+    "gpu_batches_per_second": 1.0490082691568308,
+    "producer_headroom": 10.00535214412002,
     "sustainable": true,
-    "buffer_seconds": 60.87900352039348,
+    "buffer_seconds": 60.05672390994399,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409692.50155643857,
-    "upload_bytes_per_second": 4323983.225698083,
+    "generation_bytes_per_second": 402564.9410295652,
+    "upload_bytes_per_second": 5735685.118812759,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.91047368068568,
+    "estimated_full_shards_per_hour": 86.38106511273591,
     "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": 1.8616032419995463,
-      "bytes": 11914182,
-      "pages": 32,
-      "shards": 3
-    },
-    {
-      "seconds": 1.137752229000398,
-      "bytes": 7686898,
-      "pages": 38,
-      "shards": 3
-    },
-    {
-      "seconds": 1.062077575999865,
-      "bytes": 5281467,
-      "pages": 34,
+      "seconds": 1.1886329499993735,
+      "bytes": 6263136,
+      "pages": 35,
       "shards": 2
     },
     {
-      "seconds": 0.9683684410001661,
-      "bytes": 6284844,
-      "pages": 36,
+      "seconds": 0.9854559350005729,
+      "bytes": 2421861,
+      "pages": 30,
       "shards": 2
     },
     {
-      "seconds": 0.9990272300001379,
-      "bytes": 2457257,
-      "pages": 27,
+      "seconds": 1.444283180999264,
+      "bytes": 6519870,
+      "pages": 22,
       "shards": 2
     },
     {
-      "seconds": 1.5623461810000663,
-      "bytes": 6316246,
+      "seconds": 1.324441224000111,
+      "bytes": 10689415,
       "pages": 37,
-      "shards": 2
+      "shards": 3
     },
     {
-      "seconds": 1.5784334970003329,
-      "bytes": 16058487,
-      "pages": 41,
-      "shards": 4
+      "seconds": 1.1898289029995794,
+      "bytes": 7719194,
+      "pages": 39,
+      "shards": 3
     },
     {
-      "seconds": 1.192334227000174,
-      "bytes": 2321455,
-      "pages": 29,
-      "shards": 2
+      "seconds": 0.9053547809999145,
+      "bytes": 813842,
+      "pages": 25,
+      "shards": 1
     },
     {
-      "seconds": 1.0339759600001344,
-      "bytes": 6174564,
-      "pages": 37,
-      "shards": 2
+      "seconds": 1.1792037220002385,
+      "bytes": 7645453,
+      "pages": 39,
+      "shards": 3
     },
     {
-      "seconds": 1.00537013399844,
-      "bytes": 7625189,
-      "pages": 38,
+      "seconds": 1.0877346289998968,
+      "bytes": 10017417,
+      "pages": 33,
       "shards": 3
     },
     {
-      "seconds": 1.1632388540001557,
-      "bytes": 4899738,
-      "pages": 24,
-      "shards": 2
+      "seconds": 1.1130377059998864,
+      "bytes": 986874,
+      "pages": 29,
+      "shards": 1
     },
     {
-      "seconds": 1.1982349889985926,
-      "bytes": 6456251,
-      "pages": 39,
-      "shards": 2
+      "seconds": 1.09502404000159,
+      "bytes": 13537871,
+      "pages": 40,
+      "shards": 4
     },
     {
-      "seconds": 1.1799431950003054,
-      "bytes": 8208196,
-      "pages": 35,
-      "shards": 3
+      "seconds": 0.918569364999712,
+      "bytes": 999510,
+      "pages": 30,
+      "shards": 1
     },
     {
-      "seconds": 1.0678595420013153,
-      "bytes": 6041574,
+      "seconds": 1.372887202000129,
+      "bytes": 11719387,
+      "pages": 39,
+      "shards": 4
+    },
+    {
+      "seconds": 1.0358149310013687,
+      "bytes": 6338461,
       "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 1.7238812670002517,
-      "bytes": 11592714,
-      "pages": 34,
-      "shards": 4
+      "seconds": 0.8834309400008351,
+      "bytes": 840516,
+      "pages": 27,
+      "shards": 1
     },
     {
-      "seconds": 1.1851188379987434,
-      "bytes": 1113462,
-      "pages": 32,
-      "shards": 1
+      "seconds": 1.1130535650008824,
+      "bytes": 11588700,
+      "pages": 34,
+      "shards": 4
     },
     {
-      "seconds": 1.064678984999773,
-      "bytes": 6185154,
-      "pages": 37,
+      "seconds": 1.0404952170010802,
+      "bytes": 6993608,
+      "pages": 35,
       "shards": 2
     },
     {
-      "seconds": 1.4897821249996923,
-      "bytes": 7479884,
+      "seconds": 1.010654464000254,
+      "bytes": 6096042,
       "pages": 37,
       "shards": 3
     },
     {
-      "seconds": 1.613565400999505,
-      "bytes": 11743615,
-      "pages": 37,
+      "seconds": 1.1592712370002118,
+      "bytes": 7445713,
+      "pages": 36,
       "shards": 3
     },
     {
-      "seconds": 1.0746530969991,
-      "bytes": 5830095,
-      "pages": 27,
+      "seconds": 1.0245760489997338,
+      "bytes": 5205900,
+      "pages": 31,
       "shards": 2
     },
     {
-      "seconds": 1.0229502059992228,
-      "bytes": 2562647,
-      "pages": 37,
-      "shards": 2
+      "seconds": 1.034200502999738,
+      "bytes": 7563521,
+      "pages": 38,
+      "shards": 3
     },
     {
-      "seconds": 1.0365147789998446,
-      "bytes": 6395327,
-      "pages": 37,
+      "seconds": 1.312004507999518,
+      "bytes": 5834009,
+      "pages": 26,
       "shards": 2
     },
     {
-      "seconds": 1.1455854699997872,
-      "bytes": 11238485,
-      "pages": 32,
-      "shards": 4
+      "seconds": 1.3078732329995546,
+      "bytes": 7188440,
+      "pages": 42,
+      "shards": 2
     },
     {
-      "seconds": 0.9324971470014134,
-      "bytes": 1088704,
+      "seconds": 1.37187312900096,
+      "bytes": 11914910,
       "pages": 34,
-      "shards": 1
+      "shards": 4
     },
     {
-      "seconds": 1.2869174759998714,
-      "bytes": 6108290,
-      "pages": 9,
+      "seconds": 1.1714240479996079,
+      "bytes": 6274938,
+      "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.1298799779997353,
-      "bytes": 7635963,
-      "pages": 39,
-      "shards": 4
+      "seconds": 1.1267517289998068,
+      "bytes": 2389323,
+      "pages": 28,
+      "shards": 2
     },
     {
-      "seconds": 1.3202205550005601,
-      "bytes": 6021371,
-      "pages": 27,
+      "seconds": 1.253374073001396,
+      "bytes": 6015964,
+      "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.0867800260002696,
-      "bytes": 6181304,
-      "pages": 41,
+      "seconds": 1.0633229350005422,
+      "bytes": 4861789,
+      "pages": 27,
       "shards": 2
     },
     {
-      "seconds": 25.14463806799904,
-      "bytes": 7664241,
-      "pages": 38,
-      "shards": 3
+      "seconds": 1.667441460000191,
+      "bytes": 13590477,
+      "pages": 39,
+      "shards": 4
     },
     {
-      "seconds": 1.7939357480008766,
-      "bytes": 6243705,
+      "seconds": 1.1102066319999722,
+      "bytes": 6253510,
       "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 1.0855320710015803,
-      "bytes": 8387056,
-      "pages": 37,
-      "shards": 3
+      "seconds": 1.0856502950009599,
+      "bytes": 2667766,
+      "pages": 38,
+      "shards": 2
     },
     {
-      "seconds": 1.502407841000604,
-      "bytes": 9864729,
+      "seconds": 1.1392225029994734,
+      "bytes": 10365382,
       "pages": 33,
       "shards": 3
     },
     {
-      "seconds": 1.077067022999472,
-      "bytes": 1093074,
+      "seconds": 0.9051334899995709,
+      "bytes": 1149324,
       "pages": 34,
       "shards": 1
     },
     {
-      "seconds": 1.176304970000274,
-      "bytes": 7496411,
-      "pages": 39,
+      "seconds": 1.495498950000183,
+      "bytes": 7289548,
+      "pages": 26,
       "shards": 3
     },
     {
-      "seconds": 1.0410123799993016,
-      "bytes": 5644158,
-      "pages": 19,
-      "shards": 2
+      "seconds": 1.2270069759997568,
+      "bytes": 11800938,
+      "pages": 38,
+      "shards": 3
     },
     {
-      "seconds": 1.434721014000388,
-      "bytes": 6921105,
-      "pages": 46,
-      "shards": 3
+      "seconds": 1.0093337090002024,
+      "bytes": 5575178,
+      "pages": 32,
+      "shards": 2
     },
     {
-      "seconds": 1.2365994310002861,
-      "bytes": 11662593,
-      "pages": 31,
+      "seconds": 1.1225266829987959,
+      "bytes": 7591094,
+      "pages": 38,
       "shards": 3
     },
     {
-      "seconds": 1.2687110730003042,
-      "bytes": 6191110,
+      "seconds": 1.4399703239996597,
+      "bytes": 6008739,
       "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 1.3943383940004423,
-      "bytes": 6935244,
+      "seconds": 1.3038473750020785,
+      "bytes": 2307657,
+      "pages": 29,
+      "shards": 2
+    },
+    {
+      "seconds": 1.5242876569973305,
+      "bytes": 9910776,
       "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 1.2232918650006468,
-      "bytes": 6107930,
-      "pages": 37,
+      "seconds": 0.9998362250007631,
+      "bytes": 6697585,
+      "pages": 27,
       "shards": 2
     },
     {
-      "seconds": 1.1886329499993735,
-      "bytes": 6263136,
-      "pages": 35,
-      "shards": 2
+      "seconds": 1.1621864580010879,
+      "bytes": 7532043,
+      "pages": 39,
+      "shards": 3
     },
     {
-      "seconds": 0.9854559350005729,
-      "bytes": 2421861,
-      "pages": 30,
+      "seconds": 1.2297442769995541,
+      "bytes": 6251173,
+      "pages": 35,
       "shards": 2
     },
     {
-      "seconds": 1.444283180999264,
-      "bytes": 6519870,
-      "pages": 22,
+      "seconds": 0.9681137329971534,
+      "bytes": 5823551,
+      "pages": 34,
       "shards": 2
     },
     {
-      "seconds": 1.324441224000111,
-      "bytes": 10689415,
-      "pages": 37,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1898289029995794,
-      "bytes": 7719194,
-      "pages": 39,
+      "seconds": 1.3083030150010018,
+      "bytes": 7346473,
+      "pages": 31,
       "shards": 3
     },
     {
-      "seconds": 0.9053547809999145,
-      "bytes": 813842,
-      "pages": 25,
+      "seconds": 0.8391288720013108,
+      "bytes": 1133832,
+      "pages": 34,
       "shards": 1
     },
     {
-      "seconds": 1.1792037220002385,
-      "bytes": 7645453,
-      "pages": 39,
-      "shards": 3
+      "seconds": 1.679156835001777,
+      "bytes": 13329088,
+      "pages": 34,
+      "shards": 4
     },
     {
-      "seconds": 1.0877346289998968,
-      "bytes": 10017417,
-      "pages": 33,
+      "seconds": 1.1372946299998148,
+      "bytes": 9925526,
+      "pages": 35,
       "shards": 3
     },
     {
-      "seconds": 1.1130377059998864,
-      "bytes": 986874,
-      "pages": 29,
-      "shards": 1
+      "seconds": 1.0260383269996964,
+      "bytes": 1162438,
+      "pages": 35,
+      "shards": 2
     },
     {
-      "seconds": 1.09502404000159,
-      "bytes": 13537871,
-      "pages": 40,
-      "shards": 4
+      "seconds": 1.0241040929977316,
+      "bytes": 7727150,
+      "pages": 39,
+      "shards": 3
     },
     {
-      "seconds": 0.918569364999712,
-      "bytes": 999510,
-      "pages": 30,
+      "seconds": 1.1232291689993872,
+      "bytes": 903046,
+      "pages": 26,
       "shards": 1
     },
     {
-      "seconds": 1.372887202000129,
-      "bytes": 11719387,
-      "pages": 39,
-      "shards": 4
+      "seconds": 1.245589108999411,
+      "bytes": 10344850,
+      "pages": 37,
+      "shards": 3
     },
     {
-      "seconds": 1.0358149310013687,
-      "bytes": 6338461,
-      "pages": 36,
-      "shards": 2
+      "seconds": 1.3122096120023343,
+      "bytes": 13394029,
+      "pages": 40,
+      "shards": 4
     },
     {
-      "seconds": 0.8834309400008351,
-      "bytes": 840516,
-      "pages": 27,
+      "seconds": 1.2143651210026292,
+      "bytes": 938191,
+      "pages": 29,
       "shards": 1
     },
     {
-      "seconds": 1.1130535650008824,
-      "bytes": 11588700,
-      "pages": 34,
+      "seconds": 1.3705294199971831,
+      "bytes": 12626897,
+      "pages": 40,
       "shards": 4
     },
     {
-      "seconds": 1.0404952170010802,
-      "bytes": 6993608,
+      "seconds": 1.047090729996853,
+      "bytes": 5981571,
       "pages": 35,
       "shards": 2
     },
     {
-      "seconds": 1.010654464000254,
-      "bytes": 6096042,
-      "pages": 37,
-      "shards": 3
+      "seconds": 0.8534065610001562,
+      "bytes": 1138564,
+      "pages": 34,
+      "shards": 1
     },
     {
-      "seconds": 1.1592712370002118,
-      "bytes": 7445713,
-      "pages": 36,
+      "seconds": 1.0472652070020558,
+      "bytes": 7166131,
+      "pages": 30,
       "shards": 3
     },
     {
-      "seconds": 1.0245760489997338,
-      "bytes": 5205900,
-      "pages": 31,
+      "seconds": 1.1100580310012447,
+      "bytes": 6769711,
+      "pages": 33,
       "shards": 2
     },
     {
-      "seconds": 1.034200502999738,
-      "bytes": 7563521,
+      "seconds": 1.1388595060016087,
+      "bytes": 11609898,
       "pages": 38,
-      "shards": 3
+      "shards": 4
     },
     {
-      "seconds": 1.312004507999518,
-      "bytes": 5834009,
-      "pages": 26,
+      "seconds": 1.240830022998125,
+      "bytes": 6131822,
+      "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.3078732329995546,
-      "bytes": 7188440,
-      "pages": 42,
-      "shards": 2
+      "seconds": 1.2179747789996327,
+      "bytes": 1206466,
+      "pages": 34,
+      "shards": 1
     },
     {
-      "seconds": 1.37187312900096,
-      "bytes": 11914910,
+      "seconds": 1.7537132789984753,
+      "bytes": 11388396,
       "pages": 34,
       "shards": 4
     },
     {
-      "seconds": 1.1714240479996079,
-      "bytes": 6274938,
-      "pages": 37,
+      "seconds": 1.1776387320023787,
+      "bytes": 5920541,
+      "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2547966029997042,
+      "bytes": 8617085,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

trainer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:071781bcfaf5f9809ccf892899e3b350ca09fb10a20d4e9fa8aca6156c9779ea
-size 721768191
+oid sha256:7d0f9cff5ba25bc9ee4e96e42582ae75644b70f80f581eeb59cc1552ebe7c806
+size 721782079

training_config.json

@@ -53,8 +53,8 @@
   "cpu_threads": 4,
   "page_rows": 4096,
   "cortex_page_rows": 128,
-  "prefetch_batches": 64,
-  "prefetch_low_water": 32,
+  "prefetch_batches": 63,
+  "prefetch_low_water": 31,
   "memory_pages": 12,
   "cache_gb": 20,
   "force_rebenchmark": false,