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": 76361.71093923261,
-      "required_pages_per_second_upper_estimate": 0.005282880346356624
+      "required_token_positions_per_second": 77168.11322035352,
+      "required_pages_per_second_upper_estimate": 0.00533866912727565
     },
     {
       "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": 9545.213867404076,
-      "required_pages_per_second_upper_estimate": 0.003996381719550337
+      "required_token_positions_per_second": 9646.01415254419,
+      "required_pages_per_second_upper_estimate": 0.004038584693989188
     },
     {
       "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": 19090.427734808152,
-      "required_pages_per_second_upper_estimate": 0.00397603264872573
+      "required_token_positions_per_second": 19292.02830508838,
+      "required_pages_per_second_upper_estimate": 0.00401802073095054
     },
     {
       "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": 28635.64160221223,
-      "required_pages_per_second_upper_estimate": 0.0017166629899596035
+      "required_token_positions_per_second": 28938.04245763257,
+      "required_pages_per_second_upper_estimate": 0.0017347914595026825
     },
     {
       "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": 28635.64160221223,
-      "required_pages_per_second_upper_estimate": 0.0022262951796959145
+      "required_token_positions_per_second": 28938.04245763257,
+      "required_pages_per_second_upper_estimate": 0.0022498055160840545
     },
     {
       "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": 28635.64160221223,
-      "required_pages_per_second_upper_estimate": 0.3673072639167305
+      "required_token_positions_per_second": 28938.04245763257,
+      "required_pages_per_second_upper_estimate": 0.37118613739732137
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.11231418633565227,
-    "producer_p95_seconds": 0.33518507999542635,
-    "gpu_update_seconds": 0.9655100585510915,
-    "producer_batches_per_second": 8.903594751703833,
-    "gpu_batches_per_second": 1.0357219908207547,
-    "producer_headroom": 8.59651029003276,
+    "producer_batch_seconds": 0.09218300169499116,
+    "producer_p95_seconds": 0.3287969829980284,
+    "gpu_update_seconds": 0.9554205347676407,
+    "producer_batches_per_second": 10.847986956518643,
+    "gpu_batches_per_second": 1.0466595217604373,
+    "producer_headroom": 10.364389499149432,
     "sustainable": true,
-    "buffer_seconds": 60.82713368871877,
+    "buffer_seconds": 60.19149369036136,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265738.99062988814,
+    "generation_bytes_per_second": 272736.02276552573,
     "upload_bytes_per_second": 5081534.226044488,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.021401302075226,
+    "estimated_full_shards_per_hour": 58.522801515811246,
     "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": [

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -375,3 +375,4 @@
 {"step": 594600, "loss": 0.6324679255485535, "lr": 0.0003, "grad_norm": 0.23020920522455954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87031483910, "tokens_per_second": 150174.36969955018, "data_wait_s": 1.0929994459729642e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 594700, "loss": 0.590336263179779, "lr": 0.0003, "grad_norm": 0.17049604651955155, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87046274300, "tokens_per_second": 153911.2265669135, "data_wait_s": 1.3256998499855399e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 594800, "loss": 0.5593304634094238, "lr": 0.0003, "grad_norm": 0.1940894573405728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87061060095, "tokens_per_second": 153248.34066052848, "data_wait_s": 9.212999430019408e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 594900, "loss": 0.5605776906013489, "lr": 0.0003, "grad_norm": 0.14650906269365893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87075723036, "tokens_per_second": 152239.56984673845, "data_wait_s": 1.2960997992195189e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7af4e0a3ef4eea61ced475094db0ddbcce1ed6176b3cc48789139e5143ada12
-size 804132
+oid sha256:f2af7b97cb503d86e84984d0a0f2acaee8c54fa21454116134e7d8d801088d5b
+size 806250