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": 76231.00179231007,
-      "required_pages_per_second_upper_estimate": 0.006607510029649767
+      "required_token_positions_per_second": 75715.60182369396,
+      "required_pages_per_second_upper_estimate": 0.0065628364666394
     },
     {
       "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": 9528.875224038758,
-      "required_pages_per_second_upper_estimate": 0.0038012107962497044
+      "required_token_positions_per_second": 9464.450227961745,
+      "required_pages_per_second_upper_estimate": 0.0037755107020750536
     },
     {
       "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": 19057.750448077517,
-      "required_pages_per_second_upper_estimate": 0.004408675161210349
+      "required_token_positions_per_second": 18928.90045592349,
+      "required_pages_per_second_upper_estimate": 0.0043788679832079255
     },
     {
       "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": 28586.625672116275,
-      "required_pages_per_second_upper_estimate": 0.0017407755209982328
+      "required_token_positions_per_second": 28393.350683885237,
+      "required_pages_per_second_upper_estimate": 0.001729006088250457
     },
     {
       "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": 28586.625672116275,
-      "required_pages_per_second_upper_estimate": 0.0022006775770840987
+      "required_token_positions_per_second": 28393.350683885237,
+      "required_pages_per_second_upper_estimate": 0.002185798733470663
     },
     {
       "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": 28586.625672116275,
-      "required_pages_per_second_upper_estimate": 0.3549502175660414
+      "required_token_positions_per_second": 28393.350683885237,
+      "required_pages_per_second_upper_estimate": 0.3525503890619869
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.5385579012814077,
-    "producer_p95_seconds": 0.19541943399963202,
-    "gpu_update_seconds": 0.9671655660628802,
-    "producer_batches_per_second": 1.8568105632108798,
-    "gpu_batches_per_second": 1.0339491345528167,
-    "producer_headroom": 1.795843239439386,
+    "producer_batch_seconds": 0.11498399509382295,
+    "producer_p95_seconds": 0.20149779999883322,
+    "gpu_update_seconds": 0.9737491114668526,
+    "producer_batches_per_second": 8.696862543209033,
+    "gpu_batches_per_second": 1.0269585750826546,
+    "producer_headroom": 8.468562173999148,
     "sustainable": true,
-    "buffer_seconds": 60.93143066196145,
+    "buffer_seconds": 60.372444910944864,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 296223.0253523843,
-    "upload_bytes_per_second": 5502367.179275418,
+    "generation_bytes_per_second": 308189.7344905989,
+    "upload_bytes_per_second": 5588787.192855578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.562565521513434,
+    "estimated_full_shards_per_hour": 66.13034273184276,
     "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": [
@@ -466,6 +466,12 @@
       "bytes": 5661886,
       "pages": 23,
       "shards": 2
+    },
+    {
+      "seconds": 1.3326636759993562,
+      "bytes": 12247274,
+      "pages": 47,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -57,3 +57,4 @@
 {"step": 285900, "loss": 0.5836225003004074, "lr": 0.0003, "grad_norm": 0.1586045129972069, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41847504610, "tokens_per_second": 149271.3635139179, "data_wait_s": 1.3619999663205817e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286000, "loss": 0.6062596887350082, "lr": 0.0003, "grad_norm": 0.1471150243766587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41862184336, "tokens_per_second": 139687.25513113348, "data_wait_s": 1.1812999218818732e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 286100, "loss": 0.6028264835476875, "lr": 0.0003, "grad_norm": 0.14100641591524665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41876868380, "tokens_per_second": 150696.12758620843, "data_wait_s": 1.7432001186534762e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 286200, "loss": 0.6156712435185909, "lr": 0.0003, "grad_norm": 0.15521044479639703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41891574606, "tokens_per_second": 150345.41218694282, "data_wait_s": 1.8108001313521527e-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:aff40e4a2672d845f632d573cff545ef299b72a76168237f41f4b9e8d79fafda
-size 125910
+oid sha256:f63c4dd6cb7dbc7bca70307c29e7ac38872d9f938b30f5103a7869f2b683d963
+size 128028