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": 76093.31218222965,
-      "required_pages_per_second_upper_estimate": 0.006595575443219122
+      "required_token_positions_per_second": 75548.37368533963,
+      "required_pages_per_second_upper_estimate": 0.006548341555442688
     },
     {
       "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": 9511.664022778707,
-      "required_pages_per_second_upper_estimate": 0.0037943449907366787
+      "required_token_positions_per_second": 9443.546710667453,
+      "required_pages_per_second_upper_estimate": 0.003767171976490926
     },
     {
       "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": 19023.328045557413,
-      "required_pages_per_second_upper_estimate": 0.004400712144200916
+      "required_token_positions_per_second": 18887.093421334906,
+      "required_pages_per_second_upper_estimate": 0.004369196661534536
     },
     {
       "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": 28534.992068336123,
-      "required_pages_per_second_upper_estimate": 0.0017376313054286042
+      "required_token_positions_per_second": 28330.640132002358,
+      "required_pages_per_second_upper_estimate": 0.0017251873446576277
     },
     {
       "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": 28534.992068336123,
-      "required_pages_per_second_upper_estimate": 0.0021967026793341393
+      "required_token_positions_per_second": 28330.640132002358,
+      "required_pages_per_second_upper_estimate": 0.002180971101592801
     },
     {
       "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": 28534.992068336123,
-      "required_pages_per_second_upper_estimate": 0.35430910101364743
+      "required_token_positions_per_second": 28330.640132002358,
+      "required_pages_per_second_upper_estimate": 0.35177173388631755
     }
   ],
   "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": 0.13139196334373082,
-    "producer_p95_seconds": 0.42427949600096326,
-    "gpu_update_seconds": 0.96891563641539,
-    "producer_batches_per_second": 7.610815567036837,
-    "gpu_batches_per_second": 1.0320815996938701,
-    "producer_headroom": 7.374238208775655,
+    "producer_batch_seconds": 0.14124316669534664,
+    "producer_p95_seconds": 0.4402986479981337,
+    "gpu_update_seconds": 0.97590452849559,
+    "producer_batches_per_second": 7.079988528980961,
+    "gpu_batches_per_second": 1.0246903982929094,
+    "producer_headroom": 6.909392867129351,
     "sustainable": true,
-    "buffer_seconds": 60.072769457754184,
+    "buffer_seconds": 60.506080766726576,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307390.1839945397,
-    "upload_bytes_per_second": 6389593.466569988,
+    "generation_bytes_per_second": 270401.37842881365,
+    "upload_bytes_per_second": 6345416.31829655,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.95877780797142,
+    "estimated_full_shards_per_hour": 58.02184118889148,
     "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": 1.2494100709991471,
-      "bytes": 10781388,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9040151730005164,
       "bytes": 8396980,
@@ -562,6 +556,12 @@
       "bytes": 5914010,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 0.986767267997493,
+      "bytes": 6142895,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -396,3 +396,4 @@
 {"step": 79600, "loss": 0.6493849139660597, "lr": 0.0003, "grad_norm": 0.13829361125583936, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11649123103, "tokens_per_second": 133717.5246146714, "data_wait_s": 1.4385004760697484e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 79700, "loss": 0.6050420720130205, "lr": 0.0003, "grad_norm": 0.1384838142148072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11663822473, "tokens_per_second": 150609.7271025089, "data_wait_s": 1.0958996426779777e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 79800, "loss": 0.6834732554852962, "lr": 0.0003, "grad_norm": 0.19023999182704254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11678369579, "tokens_per_second": 139857.67292984476, "data_wait_s": 1.2934004189446568e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 79900, "loss": 0.6089620068669319, "lr": 0.0003, "grad_norm": 0.14462929289110324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11692886215, "tokens_per_second": 146590.67159233542, "data_wait_s": 1.4140998246148229e-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:d6b6cfae046132ca6154c91aa2417574a1591edcce6f15d7279c25aa167a99be
-size 848610
+oid sha256:3f05328beccdcb6f9352c2dc008ae875b25b84e8b678279330d363ee93c295d1
+size 850728