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": 75674.63897654045,
-      "required_pages_per_second_upper_estimate": 0.006559285910867521
+      "required_token_positions_per_second": 75353.21624803833,
+      "required_pages_per_second_upper_estimate": 0.006531425803399414
     },
     {
       "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": 9459.329872067556,
-      "required_pages_per_second_upper_estimate": 0.0037734681155527187
+      "required_token_positions_per_second": 9419.15203100479,
+      "required_pages_per_second_upper_estimate": 0.003757440574040526
     },
     {
       "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": 18918.65974413511,
-      "required_pages_per_second_upper_estimate": 0.0043764989747660564
+      "required_token_positions_per_second": 18838.30406200958,
+      "required_pages_per_second_upper_estimate": 0.004357910101918019
     },
     {
       "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": 28377.98961620267,
-      "required_pages_per_second_upper_estimate": 0.0017280706798218823
+      "required_token_positions_per_second": 28257.456093014374,
+      "required_pages_per_second_upper_estimate": 0.0017207308206502374
     },
     {
       "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": 28377.98961620267,
-      "required_pages_per_second_upper_estimate": 0.0021846161959583024
+      "required_token_positions_per_second": 28257.456093014374,
+      "required_pages_per_second_upper_estimate": 0.00217533719168512
     },
     {
       "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": 28377.98961620267,
-      "required_pages_per_second_upper_estimate": 0.35235965601155583
+      "required_token_positions_per_second": 28257.456093014374,
+      "required_pages_per_second_upper_estimate": 0.35086303305330935
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.14572423280469593,
-    "producer_p95_seconds": 0.39436719800141873,
-    "gpu_update_seconds": 0.9742762039850111,
-    "producer_batches_per_second": 6.862276649212013,
-    "gpu_batches_per_second": 1.026402980910108,
-    "producer_headroom": 6.6857528444892615,
+    "producer_batch_seconds": 0.09481285885919988,
+    "producer_p95_seconds": 0.3795729499979643,
+    "gpu_update_seconds": 0.9784320254799922,
+    "producer_batches_per_second": 10.547092578286579,
+    "gpu_batches_per_second": 1.022043406142013,
+    "producer_headroom": 10.319613154297931,
     "sustainable": true,
-    "buffer_seconds": 60.40512464707069,
+    "buffer_seconds": 60.662785579759515,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284568.43933168886,
-    "upload_bytes_per_second": 5336556.513494245,
+    "generation_bytes_per_second": 285047.90833474684,
+    "upload_bytes_per_second": 5287078.859485369,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.0617626663494,
+    "estimated_full_shards_per_hour": 61.164645552938495,
     "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": 1.5235254969993548,
-      "bytes": 11921244,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.0007863089995226,
       "bytes": 6027863,
@@ -562,6 +556,12 @@
       "bytes": 2819009,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1499773260002257,
+      "bytes": 6017503,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 389700, "loss": 0.6059103906154633, "lr": 0.0003, "grad_norm": 0.15986029536768465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57041425564, "tokens_per_second": 148058.45425343263, "data_wait_s": 1.1693999113049358e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 389800, "loss": 0.6095029674470425, "lr": 0.0003, "grad_norm": 0.17847424212721572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57056045820, "tokens_per_second": 148399.3938685122, "data_wait_s": 9.235998732037842e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 389900, "loss": 0.6039017792791128, "lr": 0.0003, "grad_norm": 0.16267038271425097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57070706750, "tokens_per_second": 149413.2835373003, "data_wait_s": 1.9758997950702906e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 390000, "loss": 0.6361445300281048, "lr": 0.0003, "grad_norm": 0.16592915181678689, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57085381676, "tokens_per_second": 140165.60097290107, "data_wait_s": 1.0950003343168646e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be67e0e65e5cd9e630d67f16fefc5964bfa454cf8b73c0614c82ac26103a9570
-size 744045
+oid sha256:8a6dccd20560d70bbf634de5de547f2a53a5184aeab90bd0f1e7af90a20f7b13
+size 746163