Qwen90M step 445000; complete resumable checkpoint

complete.json

@@ -1,18 +1,18 @@
 {
-  "step": 440277,
+  "step": 445000,
   "files": {
     "config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7",
     "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4",
-    "model.safetensors": "8a7a87fe75a5fd44cbeb174e25a756396db70470c62d2ac3cab5cb0c7ca63f33",
+    "model.safetensors": "e31fd28a60f36256443cfb50640d20094f71da6c3cbb0aa80afc0a4cedec13b2",
     "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03",
     "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300",
     "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263",
     "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6",
-    "trainer.pt": "4647e9f281860ad757cb073bc1c37c925e1bf8c1a3363bc3857ab396fe20b7fc",
+    "trainer.pt": "8b5075127ef2811f3ef44f7a6a9a1feed435ce847072f56baa6aec7dfad54d50",
     "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6",
-    "performance.json": "a6ae17d5ca4d76a29c6e069dddb352125dfbf6f18e59e9553ab94c741a962b18",
-    "shard_scheduler.json": "5c5eb10191a168242e046d479a37b48f91a5de39c8e14d82d54a83d47dbceb69",
-    "coverage.json": "f9927ca485baa62a65235b140994e7e3f62d2fca6bbf7433fe52e93b3eef7c89",
+    "performance.json": "58f8b6f79b191aa11b2bedb4594ca2897b6fdad2419ea7c9210688ac8022e7cb",
+    "shard_scheduler.json": "efbac20cd8a66e3c031c9dfa928f6451599605e8f1347de2378f5b7118ef7df7",
+    "coverage.json": "5537feb0d0fc33dea31497140125b57b887abc964d22bb91f1d5ee833a726e2b",
     "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"
   },
   "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063"

coverage.json

@@ -1,107 +1,107 @@
 {
-  "step": 440277,
+  "step": 445000,
   "stats": {
-    "climbmix/blocks": 15849972,
-    "climbmix/targets": 26495964044,
-    "climbmix/input_tokens": 26497642217,
-    "ultra_qa/blocks": 5943739,
-    "ultra_qa/targets": 9696006191,
-    "ultra_qa/input_tokens": 9696024796,
-    "cortex/blocks": 5943740,
-    "cortex/targets": 226052219,
-    "cortex/input_tokens": 9448331338,
-    "ultra_style/blocks": 5943740,
-    "ultra_style/targets": 9401027313,
-    "ultra_style/input_tokens": 9401520126,
-    "openmath/blocks": 3962493,
-    "openmath/targets": 6122914510,
-    "openmath/input_tokens": 6124954796,
-    "rewrite6/blocks": 1981246,
-    "rewrite6/targets": 3247923015,
-    "rewrite6/input_tokens": 3251484097
+    "climbmix/blocks": 16020000,
+    "climbmix/targets": 26779783410,
+    "climbmix/input_tokens": 26781479575,
+    "ultra_qa/blocks": 6007500,
+    "ultra_qa/targets": 9800065187,
+    "ultra_qa/input_tokens": 9800083982,
+    "cortex/blocks": 6007500,
+    "cortex/targets": 228478974,
+    "cortex/input_tokens": 9549691996,
+    "ultra_style/blocks": 6007500,
+    "ultra_style/targets": 9501732854,
+    "ultra_style/input_tokens": 9502231083,
+    "openmath/blocks": 4005000,
+    "openmath/targets": 6188564834,
+    "openmath/input_tokens": 6190627055,
+    "rewrite6/blocks": 2002500,
+    "rewrite6/targets": 3282863241,
+    "rewrite6/input_tokens": 3286462391
   },
   "sources": {
     "climbmix": {
-      "page": 2170,
-      "row": 2226,
+      "page": 2193,
+      "row": 3524,
       "start": {
-        "file": 105,
+        "file": 106,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 4,
+          "row_group": 13,
           "row": 1024
         }
       },
       "epoch": 0,
       "offset": 2048,
-      "documents": 8890546
+      "documents": 8986052
     },
     "rewrite6": {
-      "page": 35,
-      "row": 249,
+      "page": 48,
+      "row": 1213,
       "start": {
         "file": 0,
         "reader": {
           "format": "jsonl-row-v1",
-          "row": 143360
+          "row": 196608
         }
       },
       "epoch": 24,
-      "offset": 2048,
-      "documents": 5073305
+      "offset": 0,
+      "documents": 5127517
     },
     "openmath": {
-      "page": 1356,
-      "row": 2756,
+      "page": 1371,
+      "row": 1084,
       "start": {
         "file": 12,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 314,
-          "row": 376
+          "row_group": 375,
+          "row": 816
         }
       },
       "epoch": 0,
       "offset": 0,
-      "documents": 5556932
+      "documents": 5616700
     },
     "ultra_qa": {
-      "page": 586,
-      "row": 2357,
+      "page": 592,
+      "row": 3518,
       "start": {
         "file": 4,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 4,
-          "row": 63470
+          "row_group": 5,
+          "row": 23370
         }
       },
       "epoch": 0,
       "offset": 2048,
-      "documents": 2402613
+      "documents": 2428350
     },
     "ultra_style": {
-      "page": 895,
-      "row": 1391,
+      "page": 905,
+      "row": 84,
       "start": {
         "file": 5,
         "reader": {
           "format": "parquet-row-v1",
-          "row_group": 2,
-          "row": 70225
+          "row_group": 3,
+          "row": 25655
         }
       },
       "epoch": 0,
-      "offset": 2048,
-      "documents": 3667311
+      "offset": 0,
+      "documents": 3706964
     },
     "cortex": {
-      "page": 115279,
-      "row": 76,
+      "page": 116515,
+      "row": 115,
       "start": null,
       "epoch": 0,
       "offset": 0,
-      "documents": 14755788
+      "documents": 14914035
     }
   }
 }

latest.json

@@ -1 +1 @@
-{"step": 440277, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "8a7a87fe75a5fd44cbeb174e25a756396db70470c62d2ac3cab5cb0c7ca63f33", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "4647e9f281860ad757cb073bc1c37c925e1bf8c1a3363bc3857ab396fe20b7fc", "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6", "performance.json": "a6ae17d5ca4d76a29c6e069dddb352125dfbf6f18e59e9553ab94c741a962b18", "shard_scheduler.json": "5c5eb10191a168242e046d479a37b48f91a5de39c8e14d82d54a83d47dbceb69", "coverage.json": "f9927ca485baa62a65235b140994e7e3f62d2fca6bbf7433fe52e93b3eef7c89", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000440277"}
+{"step": 445000, "files": {"config.json": "cbc72137da3dae65085a2a8c39ad7634d16bf92d785ca00d283ee82fe32b37e7", "generation_config.json": "54ed0d19f6226888c91157964ede8a892bd5ae175bfad85336152b3d9b6cd7c4", "model.safetensors": "e31fd28a60f36256443cfb50640d20094f71da6c3cbb0aa80afc0a4cedec13b2", "tokenization_byte.py": "279b9990f77b1eaedc7c1184fe13d60273a713ead856200f1fb9ae323c7b1e03", "tokenizer_config.json": "0ecbd4fd39e35125800d806af981ba0cacd8a7ba5f02c9e729c4ecc54af6d300", "special_tokens_map.json": "20d4b94d0e657c3f44d9123522c106e7315b6c1a1b6d259478f17bd2918b1263", "byte_vocab.json": "57a1cd2e70dd5b7cc6fd2eaa12ebfcd173c1620fa5d45c2c6282511c29b6a5f6", "trainer.pt": "8b5075127ef2811f3ef44f7a6a9a1feed435ce847072f56baa6aec7dfad54d50", "training_config.json": "c8bdf052510c2d9092e5a6824b119dd6584dc7697f8c4a8e9b7ecf3057c5bbb6", "performance.json": "58f8b6f79b191aa11b2bedb4594ca2897b6fdad2419ea7c9210688ac8022e7cb", "shard_scheduler.json": "efbac20cd8a66e3c031c9dfa928f6451599605e8f1347de2378f5b7118ef7df7", "coverage.json": "5537feb0d0fc33dea31497140125b57b887abc964d22bb91f1d5ee833a726e2b", "README.md": "9ce0f9efacbc46177a7e818b6aed274d84e7145ff944b5c7d6ecd5d33b9d4419"}, "signature": "9fee47918905b0750bee87876aa424f021ccc4507ee8db22a11d01fbc6109063", "folder": "step-000000445000"}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a7a87fe75a5fd44cbeb174e25a756396db70470c62d2ac3cab5cb0c7ca63f33
+oid sha256:e31fd28a60f36256443cfb50640d20094f71da6c3cbb0aa80afc0a4cedec13b2
 size 360622744

performance.json

@@ -44,16 +44,16 @@
     "source": "e6d135a4487a9fd1907e12903cda79e75fd61305fad149fda7ec5281279dc18b",
     "requested_microbatch_ceiling": 90
   },
-  "cache_reused": false,
+  "cache_reused": true,
   "attention": "PyTorch FlashAttention (forced; forward/backward verified)",
   "selected": {
     "microbatch": 90,
     "mode": "compiled",
-    "microbatch_seconds": 0.9302335780000135,
-    "peak_gb": 89.3772840499878,
+    "microbatch_seconds": 0.9383984099999907,
+    "peak_gb": 89.37790203094482,
     "fits_budget": true,
-    "estimated_update_seconds": 0.9302335780000135,
-    "input_tokens_per_second": 198143.7827650609
+    "estimated_update_seconds": 0.9383984099999907,
+    "input_tokens_per_second": 196419.77014858948
   },
   "measurements": [
     {
@@ -218,5 +218,6 @@
   "selection_policy": "max_batch",
   "budget_gb": 92.52170804977416,
   "reserve_gb": 2.1791439056396484,
-  "note": "Forward/backward throughput; excludes optimizer, data, logging and checkpoint time."
+  "note": "Forward/backward throughput; excludes optimizer, data, logging and checkpoint time.",
+  "legacy_seed_verified": false
 }

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": 77008.13474038464,
-      "required_pages_per_second_upper_estimate": 0.005327601444830188
+      "required_token_positions_per_second": 77280.73115412582,
+      "required_pages_per_second_upper_estimate": 0.0053464602972436255
     },
     {
       "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": 9626.01684254808,
-      "required_pages_per_second_upper_estimate": 0.004030212237885135
+      "required_token_positions_per_second": 9660.091394265728,
+      "required_pages_per_second_upper_estimate": 0.00404447854113176
     },
     {
       "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": 19252.03368509616,
-      "required_pages_per_second_upper_estimate": 0.004009690906335218
+      "required_token_positions_per_second": 19320.182788531456,
+      "required_pages_per_second_upper_estimate": 0.0040238845673680745
     },
     {
       "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": 28878.05052764424,
-      "required_pages_per_second_upper_estimate": 0.0017311950349022554
+      "required_token_positions_per_second": 28980.274182797188,
+      "required_pages_per_second_upper_estimate": 0.0017373231869421907
     },
     {
       "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": 28878.05052764424,
-      "required_pages_per_second_upper_estimate": 0.0022451414074040744
+      "required_token_positions_per_second": 28980.274182797188,
+      "required_pages_per_second_upper_estimate": 0.0022530888469578757
     },
     {
       "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": 28878.05052764424,
-      "required_pages_per_second_upper_estimate": 0.3704166253337469
+      "required_token_positions_per_second": 28980.274182797188,
+      "required_pages_per_second_upper_estimate": 0.3717278406228395
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,410 +158,242 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1179283470082737,
-    "producer_p95_seconds": 0.348671423002088,
-    "gpu_update_seconds": 0.9574053474812385,
-    "producer_batches_per_second": 8.47972540418837,
-    "gpu_batches_per_second": 1.0444896747556511,
-    "producer_headroom": 8.118534447142451,
+    "producer_batch_seconds": 0.23549360834378064,
+    "producer_p95_seconds": 0.1584173479996025,
+    "gpu_update_seconds": 0.9540282409202315,
+    "producer_batches_per_second": 4.2463997517086325,
+    "gpu_batches_per_second": 1.0481870002458473,
+    "producer_headroom": 4.051185285366694,
     "sustainable": true,
-    "buffer_seconds": 60.31653689131802,
+    "buffer_seconds": 60.10377917797459,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 282787.29616807465,
-    "upload_bytes_per_second": 5101231.701098665,
+    "generation_bytes_per_second": 486911.6116432644,
+    "upload_bytes_per_second": 3768783.117588168,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.67957080632858,
+    "estimated_full_shards_per_hour": 104.47989713643503,
     "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": 1.0408890239996254,
-      "bytes": 7081099,
-      "pages": 36,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0998482240029261,
-      "bytes": 5731633,
-      "pages": 29,
-      "shards": 2
+      "seconds": 1.0239758469999742,
+      "bytes": 22958,
+      "pages": 1,
+      "shards": 1
     },
     {
-      "seconds": 1.2273955099954037,
-      "bytes": 7571144,
-      "pages": 41,
-      "shards": 3
+      "seconds": 1.2533257390000472,
+      "bytes": 16191062,
+      "pages": 36,
+      "shards": 4
     },
     {
-      "seconds": 1.115527177003969,
-      "bytes": 5389343,
-      "pages": 30,
+      "seconds": 0.8292768119999891,
+      "bytes": 2252671,
+      "pages": 27,
       "shards": 2
     },
     {
-      "seconds": 1.2195735799978138,
-      "bytes": 6166768,
-      "pages": 38,
+      "seconds": 1.5412007810000432,
+      "bytes": 6087939,
+      "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 0.954921390999516,
-      "bytes": 2330284,
-      "pages": 28,
+      "seconds": 1.0659989769999356,
+      "bytes": 6165937,
+      "pages": 36,
       "shards": 2
     },
     {
-      "seconds": 1.4671854679982061,
-      "bytes": 11708465,
-      "pages": 38,
+      "seconds": 1.023882131999926,
+      "bytes": 6945306,
+      "pages": 30,
       "shards": 3
     },
     {
-      "seconds": 1.1728415899997344,
-      "bytes": 11022627,
-      "pages": 35,
-      "shards": 4
-    },
-    {
-      "seconds": 0.9499405899987323,
-      "bytes": 1172384,
+      "seconds": 1.1334343210000952,
+      "bytes": 6129536,
       "pages": 36,
-      "shards": 1
-    },
-    {
-      "seconds": 1.1931303429955733,
-      "bytes": 6041594,
-      "pages": 29,
       "shards": 2
     },
     {
-      "seconds": 1.0443037180011743,
-      "bytes": 7667563,
-      "pages": 39,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1781920779976645,
-      "bytes": 5828706,
+      "seconds": 1.3353518999999778,
+      "bytes": 13219093,
       "pages": 34,
-      "shards": 2
-    },
-    {
-      "seconds": 1.4654938949970528,
-      "bytes": 13087295,
-      "pages": 32,
       "shards": 4
     },
     {
-      "seconds": 1.2222818860027473,
-      "bytes": 6269271,
-      "pages": 38,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2005412010039436,
-      "bytes": 962816,
-      "pages": 30,
-      "shards": 1
-    },
-    {
-      "seconds": 1.2577394740001182,
-      "bytes": 11361644,
+      "seconds": 0.8045400529999824,
+      "bytes": 1220702,
       "pages": 34,
-      "shards": 4
-    },
-    {
-      "seconds": 0.925975751997612,
-      "bytes": 1221763,
-      "pages": 37,
       "shards": 1
     },
     {
-      "seconds": 1.0596703449991765,
-      "bytes": 5842037,
-      "pages": 31,
-      "shards": 2
-    },
-    {
-      "seconds": 1.226734026997292,
-      "bytes": 13551509,
-      "pages": 37,
-      "shards": 4
+      "seconds": 1.144875766000041,
+      "bytes": 9911722,
+      "pages": 38,
+      "shards": 3
     },
     {
-      "seconds": 1.065682766995451,
-      "bytes": 5277203,
-      "pages": 35,
+      "seconds": 0.8446817050003119,
+      "bytes": 2246420,
+      "pages": 27,
       "shards": 2
     },
     {
-      "seconds": 1.4707806820006226,
-      "bytes": 7704301,
-      "pages": 35,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0301591320021544,
-      "bytes": 1143669,
-      "pages": 32,
-      "shards": 1
+      "seconds": 1.1746671439996135,
+      "bytes": 6226498,
+      "pages": 36,
+      "shards": 2
     },
     {
-      "seconds": 2.9000736189991585,
-      "bytes": 10986050,
-      "pages": 37,
+      "seconds": 1.0685757479996028,
+      "bytes": 7863934,
+      "pages": 39,
       "shards": 3
     },
     {
-      "seconds": 1.3636105609984952,
-      "bytes": 7351661,
-      "pages": 31,
+      "seconds": 3.374404778999633,
+      "bytes": 10919751,
+      "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 1.2681014190020505,
-      "bytes": 6981210,
-      "pages": 39,
-      "shards": 3
+      "seconds": 1.3756260560003284,
+      "bytes": 6013324,
+      "pages": 36,
+      "shards": 2
     },
     {
-      "seconds": 1.0006479639996542,
-      "bytes": 7455931,
-      "pages": 31,
+      "seconds": 2.719455530999767,
+      "bytes": 7662793,
+      "pages": 30,
       "shards": 3
     },
     {
-      "seconds": 1.9897420429988415,
-      "bytes": 1194300,
+      "seconds": 0.9099277899999834,
+      "bytes": 1221459,
       "pages": 35,
       "shards": 1
     },
     {
-      "seconds": 1.1884745839997777,
-      "bytes": 9577292,
-      "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 1.1139246599996113,
-      "bytes": 7438556,
-      "pages": 40,
+      "seconds": 6.556346244999986,
+      "bytes": 10834593,
+      "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 2.768625452001288,
-      "bytes": 924425,
-      "pages": 28,
-      "shards": 1
-    },
-    {
-      "seconds": 1.3904317609994905,
-      "bytes": 18015559,
-      "pages": 40,
-      "shards": 5
-    },
-    {
-      "seconds": 0.9437920670025051,
-      "bytes": 911442,
-      "pages": 27,
-      "shards": 1
-    },
-    {
-      "seconds": 1.0681643999996595,
-      "bytes": 6436647,
-      "pages": 38,
+      "seconds": 0.8767920650002452,
+      "bytes": 2456869,
+      "pages": 35,
       "shards": 2
     },
     {
-      "seconds": 1.4995007219986292,
-      "bytes": 7398571,
-      "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 1.022340563002217,
-      "bytes": 921929,
-      "pages": 27,
-      "shards": 1
-    },
-    {
-      "seconds": 1.1844563459962956,
-      "bytes": 10179020,
-      "pages": 46,
+      "seconds": 1.450166357999933,
+      "bytes": 12214165,
+      "pages": 34,
       "shards": 3
     },
     {
-      "seconds": 1.3179188790018088,
-      "bytes": 13975639,
-      "pages": 37,
+      "seconds": 3.807216140000037,
+      "bytes": 7676406,
+      "pages": 38,
       "shards": 4
     },
     {
-      "seconds": 1.3881760859949281,
-      "bytes": 920491,
-      "pages": 27,
-      "shards": 1
-    },
-    {
-      "seconds": 1.721910915999615,
-      "bytes": 12247927,
-      "pages": 36,
-      "shards": 4
+      "seconds": 2.506013934999828,
+      "bytes": 5150101,
+      "pages": 38,
+      "shards": 2
     },
     {
-      "seconds": 1.029901437999797,
-      "bytes": 1106234,
-      "pages": 32,
-      "shards": 1
+      "seconds": 1.1954973030001383,
+      "bytes": 6009106,
+      "pages": 28,
+      "shards": 2
     },
     {
-      "seconds": 0.968032641998434,
-      "bytes": 6353702,
-      "pages": 39,
+      "seconds": 0.9665561969995906,
+      "bytes": 2658696,
+      "pages": 37,
       "shards": 2
     },
     {
-      "seconds": 1.354905072999827,
-      "bytes": 7189345,
+      "seconds": 1.0135722189997978,
+      "bytes": 5920798,
       "pages": 31,
-      "shards": 3
-    },
-    {
-      "seconds": 1.0060001309975632,
-      "bytes": 6436150,
-      "pages": 22,
       "shards": 2
     },
     {
-      "seconds": 2.564948939005262,
-      "bytes": 12088456,
-      "pages": 50,
-      "shards": 4
+      "seconds": 1.6641503069999999,
+      "bytes": 17682755,
+      "pages": 37,
+      "shards": 5
     },
     {
-      "seconds": 0.9611534259965993,
-      "bytes": 862575,
-      "pages": 27,
+      "seconds": 0.9498594410006262,
+      "bytes": 1178438,
+      "pages": 35,
       "shards": 1
     },
     {
-      "seconds": 1.0102765199990245,
-      "bytes": 6370774,
-      "pages": 38,
+      "seconds": 1.210409072999937,
+      "bytes": 6096975,
+      "pages": 35,
       "shards": 2
     },
     {
-      "seconds": 1.36733477999951,
-      "bytes": 11825285,
-      "pages": 33,
-      "shards": 4
-    },
-    {
-      "seconds": 0.7963763439984177,
-      "bytes": 1200164,
-      "pages": 36,
-      "shards": 1
-    },
-    {
-      "seconds": 4.367263744999946,
-      "bytes": 11457575,
-      "pages": 30,
+      "seconds": 1.174733095999727,
+      "bytes": 7198373,
+      "pages": 32,
       "shards": 3
     },
     {
-      "seconds": 0.8959607119977591,
-      "bytes": 2404108,
-      "pages": 30,
+      "seconds": 1.0848658659997454,
+      "bytes": 5569114,
+      "pages": 38,
       "shards": 2
     },
     {
-      "seconds": 1.3112088260022574,
-      "bytes": 10449012,
-      "pages": 40,
-      "shards": 3
-    },
-    {
-      "seconds": 1.3342199069957132,
-      "bytes": 7580590,
-      "pages": 31,
+      "seconds": 0.9946771380000428,
+      "bytes": 7343286,
+      "pages": 37,
       "shards": 3
     },
     {
-      "seconds": 1.0893147949973354,
-      "bytes": 1277647,
+      "seconds": 11.93620105500031,
+      "bytes": 11764874,
       "pages": 36,
-      "shards": 1
-    },
-    {
-      "seconds": 1.0343618879996939,
-      "bytes": 5958933,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 1.130363664997276,
-      "bytes": 7892634,
-      "pages": 23,
-      "shards": 3
-    },
-    {
-      "seconds": 1.3904960489962832,
-      "bytes": 10986012,
-      "pages": 47,
-      "shards": 4
-    },
-    {
-      "seconds": 1.050227845000336,
-      "bytes": 5898529,
-      "pages": 29,
-      "shards": 2
-    },
-    {
-      "seconds": 0.9206480690045282,
-      "bytes": 2670829,
-      "pages": 39,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0737715180002851,
-      "bytes": 5694268,
-      "pages": 23,
-      "shards": 2
-    },
-    {
-      "seconds": 1.1375886099995114,
-      "bytes": 7292178,
-      "pages": 43,
       "shards": 3
     },
     {
-      "seconds": 1.6739402300008805,
-      "bytes": 5561171,
-      "pages": 22,
+      "seconds": 0.9162838960000954,
+      "bytes": 5018838,
+      "pages": 32,
       "shards": 2
     },
     {
-      "seconds": 1.2334552800020901,
-      "bytes": 12163400,
-      "pages": 48,
+      "seconds": 1.177252060999308,
+      "bytes": 8046612,
+      "pages": 38,
       "shards": 3
     },
     {
-      "seconds": 1.5185914780013263,
-      "bytes": 255711,
-      "pages": 9,
+      "seconds": 1.0096394150004926,
+      "bytes": 1048802,
+      "pages": 30,
       "shards": 1
     },
     {
-      "seconds": 1.23066899299738,
-      "bytes": 11344147,
-      "pages": 36,
-      "shards": 4
+      "seconds": 0.9894983370004411,
+      "bytes": 7420139,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {
@@ -604,5 +436,9 @@
     "cpu_count": 20,
     "microbatch": 90
   },
-  "cache_reused": false
+  "cache_reused": true,
+  "resume_probe_seconds": [
+    2.3715912330000037,
+    0.09919368099997428
+  ]
 }

trainer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4647e9f281860ad757cb073bc1c37c925e1bf8c1a3363bc3857ab396fe20b7fc
-size 721777791
+oid sha256:8b5075127ef2811f3ef44f7a6a9a1feed435ce847072f56baa6aec7dfad54d50
+size 721771135