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": 75447.68256784721,
-      "required_pages_per_second_upper_estimate": 0.00653961390457775
+      "required_token_positions_per_second": 75645.04978238109,
+      "required_pages_per_second_upper_estimate": 0.006556721194510914
     },
     {
       "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": 9430.960320980901,
-      "required_pages_per_second_upper_estimate": 0.0037621510774616647
+      "required_token_positions_per_second": 9455.631222797636,
+      "required_pages_per_second_upper_estimate": 0.0037719926690592135
     },
     {
       "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": 18861.920641961802,
-      "required_pages_per_second_upper_estimate": 0.0043633733820523535
+      "required_token_positions_per_second": 18911.262445595272,
+      "required_pages_per_second_upper_estimate": 0.004374787739936872
     },
     {
       "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": 28292.8809629427,
-      "required_pages_per_second_upper_estimate": 0.001722888009368944
+      "required_token_positions_per_second": 28366.893668392906,
+      "required_pages_per_second_upper_estimate": 0.0017273949948162066
     },
     {
       "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": 28292.8809629427,
-      "required_pages_per_second_upper_estimate": 0.002178064296234519
+      "required_token_positions_per_second": 28366.893668392906,
+      "required_pages_per_second_upper_estimate": 0.002183761999180357
     },
     {
       "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": 28292.8809629427,
-      "required_pages_per_second_upper_estimate": 0.3513028913784062
+      "required_token_positions_per_second": 28366.893668392906,
+      "required_pages_per_second_upper_estimate": 0.352221881475507
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.13176313793732675,
-    "producer_p95_seconds": 0.31740701099988655,
-    "gpu_update_seconds": 0.9772069530922867,
-    "producer_batches_per_second": 7.58937602469403,
-    "gpu_batches_per_second": 1.0233246876064346,
-    "producer_headroom": 7.416391020962904,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.58683109172178,
+    "buffer_seconds": 60.42875261699793,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404897.4438805789,
-    "upload_bytes_per_second": 3488050.8040154255,
+    "generation_bytes_per_second": 388872.49483858666,
+    "upload_bytes_per_second": 3590812.0492285606,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.88156592667605,
+    "estimated_full_shards_per_hour": 83.44298490398597,
     "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": 3.285385278999456,
-      "bytes": 842975,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 2.201649025999359,
       "bytes": 9943938,
@@ -562,6 +556,12 @@
       "bytes": 2626843,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.2158071900012146,
+      "bytes": 5739703,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -234,3 +234,4 @@
 {"step": 303600, "loss": 0.612817732617259, "lr": 0.0003, "grad_norm": 0.1585622809536886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44439193986, "tokens_per_second": 149973.40725128085, "data_wait_s": 7.378999725915492e-06, "gpu_peak_gb": 85.46991729736328}
 {"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}

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 sha256:48c35335750bc37aa3a41b20e53b05a5de39b85c88bd40f59ef7cacced0a7e5e
-size 503145
+oid sha256:2a4841d986eaae453385d88c291052753049c940e64bf7c4ae3c89bdd21ca471
+size 505263