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": 76181.48704047244,
-      "required_pages_per_second_upper_estimate": 0.006603218216454472
+      "required_token_positions_per_second": 75082.5538451286,
+      "required_pages_per_second_upper_estimate": 0.006507965472302788
     },
     {
       "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": 9522.685880059054,
-      "required_pages_per_second_upper_estimate": 0.003798741774397261
+      "required_token_positions_per_second": 9385.319230641075,
+      "required_pages_per_second_upper_estimate": 0.0037439441641299964
     },
     {
       "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": 19045.37176011811,
-      "required_pages_per_second_upper_estimate": 0.004405811569608407
+      "required_token_positions_per_second": 18770.63846128215,
+      "required_pages_per_second_upper_estimate": 0.004342256856063627
     },
     {
       "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": 28568.057640177165,
-      "required_pages_per_second_upper_estimate": 0.0017396448252720755
+      "required_token_positions_per_second": 28155.957691923228,
+      "required_pages_per_second_upper_estimate": 0.0017145501005447418
     },
     {
       "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": 28568.057640177165,
-      "required_pages_per_second_upper_estimate": 0.0021992481585858237
+      "required_token_positions_per_second": 28155.957691923228,
+      "required_pages_per_second_upper_estimate": 0.0021675235638035665
     },
     {
       "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": 28568.057640177165,
-      "required_pages_per_second_upper_estimate": 0.354719664752563
+      "required_token_positions_per_second": 28155.957691923228,
+      "required_pages_per_second_upper_estimate": 0.3496027626050539
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.2782963426249836,
-    "producer_p95_seconds": 0.2674591969989706,
-    "gpu_update_seconds": 0.9677941828679586,
-    "producer_batches_per_second": 3.5932919224437794,
-    "gpu_batches_per_second": 1.0332775477494633,
-    "producer_headroom": 3.4775670198875135,
+    "producer_batch_seconds": 0.061135110367146694,
+    "producer_p95_seconds": 0.19654228599938506,
+    "gpu_update_seconds": 0.9819591399631582,
+    "producer_batches_per_second": 16.35721263925923,
+    "gpu_batches_per_second": 1.0183723123525472,
+    "producer_headroom": 16.062114455441495,
     "sustainable": true,
-    "buffer_seconds": 60.00323933781343,
+    "buffer_seconds": 60.88146667771581,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460033.5287919876,
+    "generation_bytes_per_second": 462376.4105052268,
     "upload_bytes_per_second": 6278213.682149331,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.71248624629709,
+    "estimated_full_shards_per_hour": 99.21521412246325,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -145,3 +145,4 @@
 {"step": 54500, "loss": 0.6466125007718801, "lr": 0.0003, "grad_norm": 0.13557349814196135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7976756607, "tokens_per_second": 150395.21269945215, "data_wait_s": 8.86999987415038e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 54600, "loss": 0.6637112759053707, "lr": 0.0003, "grad_norm": 0.14400315406123818, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7991340640, "tokens_per_second": 148949.89640637257, "data_wait_s": 1.491699913458433e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 54700, "loss": 0.6260981969535351, "lr": 0.0003, "grad_norm": 0.13590119837122686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8005840156, "tokens_per_second": 148382.28499399193, "data_wait_s": 7.453998478013091e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 54800, "loss": 0.6338471099734306, "lr": 0.0003, "grad_norm": 0.13873746194503214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8020481174, "tokens_per_second": 150470.1087460764, "data_wait_s": 1.324500044574961e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b00a059f95ed81bfce220a0767ca7ad21bcf3e7233e19e81a079cd2917d96df8
-size 313077
+oid sha256:a9c6d1cf853e63730306c79aa895fdb1adbaf374b137efdff9ed07454166853d
+size 315195