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": 76943.01952991771,
-      "required_pages_per_second_upper_estimate": 0.005323096623481982
+      "required_token_positions_per_second": 76731.553648087,
+      "required_pages_per_second_upper_estimate": 0.005308466923108487
     },
     {
       "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": 9617.877441239714,
-      "required_pages_per_second_upper_estimate": 0.004026804440527349
+      "required_token_positions_per_second": 9591.444206010876,
+      "required_pages_per_second_upper_estimate": 0.0040157373969257545
     },
     {
       "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": 19235.75488247943,
-      "required_pages_per_second_upper_estimate": 0.004006300461051046
+      "required_token_positions_per_second": 19182.88841202175,
+      "required_pages_per_second_upper_estimate": 0.0039952897694372925
     },
     {
       "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": 28853.632323719143,
-      "required_pages_per_second_upper_estimate": 0.0017297312008613841
+      "required_token_positions_per_second": 28774.332618032626,
+      "required_pages_per_second_upper_estimate": 0.001724977304589117
     },
     {
       "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": 28853.632323719143,
-      "required_pages_per_second_upper_estimate": 0.0022432429994532262
+      "required_token_positions_per_second": 28774.332618032626,
+      "required_pages_per_second_upper_estimate": 0.0022370777961386425
     },
     {
       "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": 28853.632323719143,
-      "required_pages_per_second_upper_estimate": 0.3701034148320204
+      "required_token_positions_per_second": 28774.332618032626,
+      "required_pages_per_second_upper_estimate": 0.36908624335286394
     }
   ],
   "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": 0.8214605102421473,
-    "producer_p95_seconds": 0.19235912399926747,
-    "gpu_update_seconds": 0.9582155788847405,
-    "producer_batches_per_second": 1.2173439715382344,
-    "gpu_batches_per_second": 1.0436064931900733,
-    "producer_headroom": 1.1664779583893583,
+    "producer_batch_seconds": 0.07528932042193048,
+    "producer_p95_seconds": 0.19481521699890436,
+    "gpu_update_seconds": 0.9608563425958744,
+    "producer_batches_per_second": 13.28209624414032,
+    "gpu_batches_per_second": 1.0407383036036106,
+    "producer_headroom": 12.762186419151067,
     "sustainable": true,
-    "buffer_seconds": 60.36758146973865,
+    "buffer_seconds": 60.533949583540085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 259275.64765156538,
-    "upload_bytes_per_second": 4451515.826357642,
+    "generation_bytes_per_second": 257586.03870410664,
+    "upload_bytes_per_second": 4507155.113609987,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.63451835785123,
+    "estimated_full_shards_per_hour": 55.27196760980987,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 7094094,
       "pages": 25,
       "shards": 3
+    },
+    {
+      "seconds": 1.3368760009998368,
+      "bytes": 10726853,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 533600, "loss": 0.5504591464996338, "lr": 0.0003, "grad_norm": 0.17225628327564876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78102309957, "tokens_per_second": 153145.91271492338, "data_wait_s": 9.602999853086658e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 533700, "loss": 0.5592876672744751, "lr": 0.0003, "grad_norm": 0.16260922391123797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78116873994, "tokens_per_second": 105166.14067153286, "data_wait_s": 1.436300044588279e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 533800, "loss": 0.5632269978523254, "lr": 0.0003, "grad_norm": 0.15193763187468254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78131356452, "tokens_per_second": 151070.21003535716, "data_wait_s": 1.2249000064912252e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 533900, "loss": 0.598848283290863, "lr": 0.0003, "grad_norm": 0.18830756847486924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78145929961, "tokens_per_second": 151962.76044866347, "data_wait_s": 9.510000381851569e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d7f4bd458ec2bbab1e951e383c6fc80c213a74b15a96f2094c027551bdc2b50
-size 187332
+oid sha256:5115e9b824ae611ca5a55dc9b41bf438b15ba4e316ed0c215f5774ab3071368c
+size 189450