Update TensorBoard training metrics

benchmark_cache/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": 77293.79132108188,
-      "required_pages_per_second_upper_estimate": 0.0053473638298974085
+      "required_token_positions_per_second": 77148.10039468572,
+      "required_pages_per_second_upper_estimate": 0.005337284593559799
     },
     {
       "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": 9661.723915135235,
-      "required_pages_per_second_upper_estimate": 0.00404516204352891
+      "required_token_positions_per_second": 9643.512549335715,
+      "required_pages_per_second_upper_estimate": 0.004037537324965214
     },
     {
       "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": 19323.44783027047,
-      "required_pages_per_second_upper_estimate": 0.0040245645894573705
+      "required_token_positions_per_second": 19287.02509867143,
+      "required_pages_per_second_upper_estimate": 0.004016978694997316
     },
     {
       "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": 28985.171745405703,
-      "required_pages_per_second_upper_estimate": 0.001737616788342426
+      "required_token_positions_per_second": 28930.537648007146,
+      "required_pages_per_second_upper_estimate": 0.001734341557624819
     },
     {
       "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": 28985.171745405703,
-      "required_pages_per_second_upper_estimate": 0.002253469610908587
+      "required_token_positions_per_second": 28930.537648007146,
+      "required_pages_per_second_upper_estimate": 0.0022492220501458466
     },
     {
       "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": 28985.171745405703,
-      "required_pages_per_second_upper_estimate": 0.37179066129738847
+      "required_token_positions_per_second": 28930.537648007146,
+      "required_pages_per_second_upper_estimate": 0.3710898737574832
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.14770525844519966,
-    "producer_p95_seconds": 0.24126123799942434,
-    "gpu_update_seconds": 0.9538670408044881,
-    "producer_batches_per_second": 6.770239668691358,
-    "gpu_batches_per_second": 1.0483641400971393,
-    "producer_headroom": 6.457908478311784,
+    "producer_batch_seconds": 0.06708961735918706,
+    "producer_p95_seconds": 0.23458091999782482,
+    "gpu_update_seconds": 0.9556683783892452,
+    "producer_batches_per_second": 14.905436032615304,
+    "gpu_batches_per_second": 1.0463880804400734,
+    "producer_headroom": 14.244653882474093,
     "sustainable": true,
-    "buffer_seconds": 60.09362357068275,
+    "buffer_seconds": 60.20710783852245,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 263022.05481579807,
-    "upload_bytes_per_second": 4924933.096373835,
+    "generation_bytes_per_second": 263212.16195785574,
+    "upload_bytes_per_second": 4979001.224232591,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.438410123400274,
+    "estimated_full_shards_per_hour": 56.47920269061808,
     "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": 1.185996062000413,
-      "bytes": 6230664,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9238252899995132,
       "bytes": 2424834,
@@ -562,6 +556,12 @@
       "bytes": 6784780,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.1795219960004033,
+      "bytes": 10882665,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -221,3 +221,4 @@
 {"step": 462300, "loss": 0.5710533857345581, "lr": 0.0003, "grad_norm": 0.14465696382242818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67666901277, "tokens_per_second": 153912.00691303314, "data_wait_s": 1.871300264610909e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 462400, "loss": 0.579341471195221, "lr": 0.0003, "grad_norm": 0.1633562735198258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67681619957, "tokens_per_second": 153178.49599468184, "data_wait_s": 1.148999945144169e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 462500, "loss": 0.6043637990951538, "lr": 0.0003, "grad_norm": 0.19626290176222383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67696317791, "tokens_per_second": 152813.25386393775, "data_wait_s": 8.059996616793796e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 462600, "loss": 0.5018669366836548, "lr": 0.0003, "grad_norm": 0.1580707417784199, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67710761565, "tokens_per_second": 149731.8577879642, "data_wait_s": 7.398000889224932e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fbda915a3df664f36274d7fc514083ebe32f2bef112084159dd8da7d90db73aa
-size 475611
+oid sha256:82ea5d9a2b5d7dfedfad247c767ee81ff8a110a934e0a4eecca5f42f71ce99a4
+size 477729