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": 77055.39053682695,
-      "required_pages_per_second_upper_estimate": 0.005330870710476621
+      "required_token_positions_per_second": 77340.98346738007,
+      "required_pages_per_second_upper_estimate": 0.005350628692079183
     },
     {
       "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": 9631.923817103368,
-      "required_pages_per_second_upper_estimate": 0.004032685364779779
+      "required_token_positions_per_second": 9667.622933422508,
+      "required_pages_per_second_upper_estimate": 0.0040476318393002815
     },
     {
       "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": 19263.847634206737,
-      "required_pages_per_second_upper_estimate": 0.004012151440380161
+      "required_token_positions_per_second": 19335.245866845016,
+      "required_pages_per_second_upper_estimate": 0.004027021809340701
     },
     {
       "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": 28895.771451310105,
-      "required_pages_per_second_upper_estimate": 0.0017322573772177402
+      "required_token_positions_per_second": 29002.868800267526,
+      "required_pages_per_second_upper_estimate": 0.0017386776997595542
     },
     {
       "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": 28895.771451310105,
-      "required_pages_per_second_upper_estimate": 0.0022465191312729896
+      "required_token_positions_per_second": 29002.868800267526,
+      "required_pages_per_second_upper_estimate": 0.0022548454790829757
     },
     {
       "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": 28895.771451310105,
-      "required_pages_per_second_upper_estimate": 0.3706439303152872
+      "required_token_positions_per_second": 29002.868800267526,
+      "required_pages_per_second_upper_estimate": 0.3720176601155389
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.07241884508592733,
-    "producer_p95_seconds": 0.1788760009994803,
-    "gpu_update_seconds": 0.9568181990429768,
-    "producer_batches_per_second": 13.808560448781906,
-    "gpu_batches_per_second": 1.0451306225155566,
-    "producer_headroom": 13.212281939979583,
+    "producer_batch_seconds": 0.08420161782805735,
+    "producer_p95_seconds": 0.17875256900060776,
+    "gpu_update_seconds": 0.9532850074384701,
+    "producer_batches_per_second": 11.876256368874467,
+    "gpu_batches_per_second": 1.049004224546713,
+    "producer_headroom": 11.321457140943675,
     "sustainable": true,
-    "buffer_seconds": 60.27954653970754,
+    "buffer_seconds": 60.05695546862362,
     "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": 462273.62131834397,
-    "upload_bytes_per_second": 5452492.670439877,
+    "generation_bytes_per_second": 492551.63341194217,
+    "upload_bytes_per_second": 5462584.158989212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.193157955768,
+    "estimated_full_shards_per_hour": 105.69011451500606,
     "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": [
@@ -484,6 +484,12 @@
       "bytes": 837425,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.2931989250000697,
+      "bytes": 7658868,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -59,3 +59,4 @@
 {"step": 610300, "loss": 0.6092767119407654, "lr": 0.0001, "grad_norm": 0.14803397781210065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89330207406, "tokens_per_second": 152671.29239890774, "data_wait_s": 1.3412999578577e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610400, "loss": 0.5435611009597778, "lr": 0.0001, "grad_norm": 0.15360692895727052, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89344914056, "tokens_per_second": 152009.4576179536, "data_wait_s": 1.6773999959696084e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610500, "loss": 0.6074555516242981, "lr": 0.0001, "grad_norm": 0.169522746046445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89359602774, "tokens_per_second": 152499.79723854814, "data_wait_s": 1.5468999663426075e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 610600, "loss": 0.5554715991020203, "lr": 0.0001, "grad_norm": 0.16461564484946448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89374259655, "tokens_per_second": 152961.84046897935, "data_wait_s": 1.0048000149254221e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b33d04bf708fbcd23bcd5a1898cc0adaddb6faeb4fb927f0b4e182443461e98
-size 130929
+oid sha256:fd808bd57ac3b1b8f82988d8aa43e3dbb86aed1cac391d5c1a310c25c7ee06c7
+size 133047