Update TensorBoard training metrics

benchmark_cache/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": 75964.97413821687,
-      "required_pages_per_second_upper_estimate": 0.006584451426833894
+      "required_token_positions_per_second": 76155.01078101793,
+      "required_pages_per_second_upper_estimate": 0.006600923321388422
     },
     {
       "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": 9495.621767277109,
-      "required_pages_per_second_upper_estimate": 0.0037879454951639975
+      "required_token_positions_per_second": 9519.376347627242,
+      "required_pages_per_second_upper_estimate": 0.0037974215524282915
     },
     {
       "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": 18991.243534554218,
-      "required_pages_per_second_upper_estimate": 0.004393289957248435
+      "required_token_positions_per_second": 19038.752695254483,
+      "required_pages_per_second_upper_estimate": 0.00440428036544377
     },
     {
       "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": 28486.865301831327,
-      "required_pages_per_second_upper_estimate": 0.0017347006378500916
+      "required_token_positions_per_second": 28558.129042881726,
+      "required_pages_per_second_upper_estimate": 0.0017390402257881077
     },
     {
       "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": 28486.865301831327,
-      "required_pages_per_second_upper_estimate": 0.002192997747625184
+      "required_token_positions_per_second": 28558.129042881726,
+      "required_pages_per_second_upper_estimate": 0.0021984838276819046
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28486.865301831327,
-      "required_pages_per_second_upper_estimate": 0.35371152764358404
+      "required_token_positions_per_second": 28558.129042881726,
+      "required_pages_per_second_upper_estimate": 0.35459638480303124
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.08164002452348029,
-    "producer_p95_seconds": 0.19491624799957208,
-    "gpu_update_seconds": 0.9705525584179527,
-    "producer_batches_per_second": 12.248893919824734,
-    "gpu_batches_per_second": 1.0303409035673947,
-    "producer_headroom": 11.888195331676002,
+    "producer_batch_seconds": 0.06285995994522864,
+    "producer_p95_seconds": 0.20234570899992832,
+    "gpu_update_seconds": 0.9681306488420472,
+    "producer_batches_per_second": 15.908377938377999,
+    "gpu_batches_per_second": 1.0329184404977476,
+    "producer_headroom": 15.401388255506403,
     "sustainable": true,
-    "buffer_seconds": 60.17425862191307,
+    "buffer_seconds": 60.024100228206926,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296136.3798160484,
-    "upload_bytes_per_second": 5559616.074557977,
+    "generation_bytes_per_second": 373637.31285776687,
+    "upload_bytes_per_second": 5504907.379593526,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.5439734064206,
+    "estimated_full_shards_per_hour": 80.17386950778727,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 7445606,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.7973876159994688,
+      "bytes": 1188820,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 286100, "loss": 0.6028264835476875, "lr": 0.0003, "grad_norm": 0.14100641591524665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41876868380, "tokens_per_second": 150696.12758620843, "data_wait_s": 1.7432001186534762e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286200, "loss": 0.6156712435185909, "lr": 0.0003, "grad_norm": 0.15521044479639703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41891574606, "tokens_per_second": 150345.41218694282, "data_wait_s": 1.8108001313521527e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286300, "loss": 0.5622965320944786, "lr": 0.0003, "grad_norm": 0.14726805877539556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41906283753, "tokens_per_second": 150687.9089786302, "data_wait_s": 1.4015000488143414e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 286400, "loss": 0.6382241677492857, "lr": 0.0003, "grad_norm": 0.15030694681059412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41921058959, "tokens_per_second": 151047.22960489857, "data_wait_s": 1.3300999853527173e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6d0ce4c02c172c9095783db06e9cfbcf7c368ee7620625b9bfd61bdfdae6c965
-size 130146
+oid sha256:5b10a7b24c05c1337d763a7a8e37afa807a7a4b490ab8f5de092c5a01ee2d2bf
+size 132264