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": 75271.64335470792,
-      "required_pages_per_second_upper_estimate": 0.006524355271750143
+      "required_token_positions_per_second": 75050.91323713194,
+      "required_pages_per_second_upper_estimate": 0.006505222944594025
     },
     {
       "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": 9408.95541933849,
-      "required_pages_per_second_upper_estimate": 0.0037533729931939087
+      "required_token_positions_per_second": 9381.364154641493,
+      "required_pages_per_second_upper_estimate": 0.0037423664251801073
     },
     {
       "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": 18817.91083867698,
-      "required_pages_per_second_upper_estimate": 0.0043531924886084945
+      "required_token_positions_per_second": 18762.728309282986,
+      "required_pages_per_second_upper_estimate": 0.004340426981612541
     },
     {
       "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": 28226.86625801547,
-      "required_pages_per_second_upper_estimate": 0.0017188680601912649
+      "required_token_positions_per_second": 28144.092463924477,
+      "required_pages_per_second_upper_estimate": 0.001713827570411658
     },
     {
       "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": 28226.86625801547,
-      "required_pages_per_second_upper_estimate": 0.0021729823015088213
+      "required_token_positions_per_second": 28144.092463924477,
+      "required_pages_per_second_upper_estimate": 0.0021666101457071697
     },
     {
       "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": 28226.86625801547,
-      "required_pages_per_second_upper_estimate": 0.35048320968021496
+      "required_token_positions_per_second": 28144.092463924477,
+      "required_pages_per_second_upper_estimate": 0.34945543618367303
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.22757804343746102,
-    "producer_p95_seconds": 0.3193292100004328,
-    "gpu_update_seconds": 0.9794923654392172,
-    "producer_batches_per_second": 4.394097009076371,
-    "gpu_batches_per_second": 1.020937002966416,
-    "producer_headroom": 4.303984473389605,
+    "producer_batch_seconds": 0.09538216467964844,
+    "producer_p95_seconds": 0.2938880830006383,
+    "gpu_update_seconds": 0.9823731227233432,
+    "producer_batches_per_second": 10.484140335445423,
+    "gpu_batches_per_second": 1.017943159140787,
+    "producer_headroom": 10.29933768040128,
     "sustainable": true,
-    "buffer_seconds": 60.72852665723147,
+    "buffer_seconds": 60.90713360884728,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 527412.8980385693,
-    "upload_bytes_per_second": 5170516.054552246,
+    "generation_bytes_per_second": 547035.6375764761,
+    "upload_bytes_per_second": 5189441.022970066,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 113.17053037517364,
+    "estimated_full_shards_per_hour": 117.38111348601066,
     "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": 1.1743630839991965,
-      "bytes": 5760431,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.9926189179968787,
       "bytes": 7099055,
@@ -562,6 +556,12 @@
       "bytes": 6643361,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8856843470020976,
+      "bytes": 5828359,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -239,3 +239,4 @@
 {"step": 63900, "loss": 0.6261835917830467, "lr": 0.0003, "grad_norm": 0.13677463701096396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9351811507, "tokens_per_second": 147353.44496967868, "data_wait_s": 1.4470002497546375e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 64000, "loss": 0.6302631683647633, "lr": 0.0003, "grad_norm": 0.13780771526704533, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9366397123, "tokens_per_second": 148638.90808606657, "data_wait_s": 1.1954001820413396e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 64100, "loss": 0.5892891921103001, "lr": 0.0003, "grad_norm": 0.1285240648130058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9380929376, "tokens_per_second": 148404.10053383745, "data_wait_s": 1.1047002772102132e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 64200, "loss": 0.6446213405579329, "lr": 0.0003, "grad_norm": 0.1374295009308241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9395625453, "tokens_per_second": 150108.32928739092, "data_wait_s": 8.966002496890724e-06, "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 sha256:612722e68d0d09611c0ccd6e0311c45ebebb5864bf384a3a7244d304bc9e473c
-size 513735
+oid sha256:42b4511264f5c738ef55625c6e70247732371b725e8681e288fd0d634d31c8f6
+size 515853