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": 75645.04978238109,
-      "required_pages_per_second_upper_estimate": 0.006556721194510914
+      "required_token_positions_per_second": 75266.12783729674,
+      "required_pages_per_second_upper_estimate": 0.006523877200679893
     },
     {
       "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": 9455.631222797636,
-      "required_pages_per_second_upper_estimate": 0.0037719926690592135
+      "required_token_positions_per_second": 9408.265979662092,
+      "required_pages_per_second_upper_estimate": 0.0037530979653989517
     },
     {
       "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": 18911.262445595272,
-      "required_pages_per_second_upper_estimate": 0.004374787739936872
+      "required_token_positions_per_second": 18816.531959324184,
+      "required_pages_per_second_upper_estimate": 0.004352873509137672
     },
     {
       "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": 28366.893668392906,
-      "required_pages_per_second_upper_estimate": 0.0017273949948162066
+      "required_token_positions_per_second": 28224.797938986278,
+      "required_pages_per_second_upper_estimate": 0.0017187421104140183
     },
     {
       "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": 28366.893668392906,
-      "required_pages_per_second_upper_estimate": 0.002183761999180357
+      "required_token_positions_per_second": 28224.797938986278,
+      "required_pages_per_second_upper_estimate": 0.002172823076584481
     },
     {
       "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": 28366.893668392906,
-      "required_pages_per_second_upper_estimate": 0.352221881475507
+      "required_token_positions_per_second": 28224.797938986278,
+      "required_pages_per_second_upper_estimate": 0.35045752808009084
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09989709145321513,
-    "producer_p95_seconds": 0.3384495770005742,
-    "gpu_update_seconds": 0.9746573002741602,
-    "producer_batches_per_second": 10.010301455756904,
-    "gpu_batches_per_second": 1.0260016517792574,
-    "producer_headroom": 9.75661339179852,
+    "producer_batch_seconds": 0.401112554195322,
+    "producer_p95_seconds": 0.35008929000105127,
+    "gpu_update_seconds": 0.9795641428422927,
+    "producer_batches_per_second": 2.4930658228988003,
+    "gpu_batches_per_second": 1.0208621939737512,
+    "producer_headroom": 2.4421178858572783,
     "sustainable": true,
-    "buffer_seconds": 60.42875261699793,
+    "buffer_seconds": 60.732976856222145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 388872.49483858666,
+    "generation_bytes_per_second": 363411.8528995383,
     "upload_bytes_per_second": 3590812.0492285606,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.44298490398597,
+    "estimated_full_shards_per_hour": 77.97972383727657,
     "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-v1/1790703713325746110-adad6d18/train.jsonl

@@ -235,3 +235,4 @@
 {"step": 303700, "loss": 0.6169750634580851, "lr": 0.0003, "grad_norm": 0.18495599045439626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44453827157, "tokens_per_second": 150129.53357710873, "data_wait_s": 1.8739003280643374e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303800, "loss": 0.5487282816320658, "lr": 0.0003, "grad_norm": 0.14358388769410568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44468555021, "tokens_per_second": 150773.0917221522, "data_wait_s": 1.5060999430716038e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303900, "loss": 0.6230304967612028, "lr": 0.0003, "grad_norm": 0.16112688015006368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44483237741, "tokens_per_second": 150450.92052689544, "data_wait_s": 1.2651002180064097e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 304000, "loss": 0.5920179393142462, "lr": 0.0003, "grad_norm": 0.15579633301433685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44497869499, "tokens_per_second": 149782.44256119436, "data_wait_s": 8.88900103745982e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2a4841d986eaae453385d88c291052753049c940e64bf7c4ae3c89bdd21ca471
-size 505263
+oid sha256:251a596a2ec03af75fccc15d68dec117971b640a0a483808f8ffaf5dd7cc5806
+size 507381