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": 76075.13267074381,
-      "required_pages_per_second_upper_estimate": 0.00659399968923908
+      "required_token_positions_per_second": 76326.2032594022,
+      "required_pages_per_second_upper_estimate": 0.0066157618515314005
     },
     {
       "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": 9509.391583842977,
-      "required_pages_per_second_upper_estimate": 0.0037934384808692265
+      "required_token_positions_per_second": 9540.775407425275,
+      "required_pages_per_second_upper_estimate": 0.0038059579573261827
     },
     {
       "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": 19018.783167685953,
-      "required_pages_per_second_upper_estimate": 0.004399660766692472
+      "required_token_positions_per_second": 19081.55081485055,
+      "required_pages_per_second_upper_estimate": 0.004414180957191142
     },
     {
       "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": 28528.17475152893,
-      "required_pages_per_second_upper_estimate": 0.0017372161666027433
+      "required_token_positions_per_second": 28622.326222275828,
+      "required_pages_per_second_upper_estimate": 0.0017429494971966386
     },
     {
       "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": 28528.17475152893,
-      "required_pages_per_second_upper_estimate": 0.002196177863414781
+      "required_token_positions_per_second": 28622.326222275828,
+      "required_pages_per_second_upper_estimate": 0.002203425904260829
     },
     {
       "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": 28528.17475152893,
-      "required_pages_per_second_upper_estimate": 0.35422445275499376
+      "required_token_positions_per_second": 28622.326222275828,
+      "required_pages_per_second_upper_estimate": 0.3553934989169677
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14040043821105996,
-    "producer_p95_seconds": 0.28223465400151326,
-    "gpu_update_seconds": 0.9691471761094088,
-    "producer_batches_per_second": 7.122484892082236,
-    "gpu_batches_per_second": 1.0318350242885175,
-    "producer_headroom": 6.902736120043426,
+    "producer_batch_seconds": 0.10175217656248492,
+    "producer_p95_seconds": 0.25219360800110735,
+    "gpu_update_seconds": 0.9659592230656102,
+    "producer_batches_per_second": 9.827799598821464,
+    "gpu_batches_per_second": 1.035240387090416,
+    "producer_headroom": 9.493253664922097,
     "sustainable": true,
-    "buffer_seconds": 60.087124918783346,
+    "buffer_seconds": 60.855431053133444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 430417.29008097673,
-    "upload_bytes_per_second": 3972863.3572530756,
+    "generation_bytes_per_second": 430442.9643781297,
+    "upload_bytes_per_second": 3934725.340146618,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.35753084966638,
+    "estimated_full_shards_per_hour": 92.36303995616835,
     "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": 1.3723258429999987,
-      "bytes": 11921886,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.34920726099881,
       "bytes": 7816018,
@@ -562,6 +556,12 @@
       "bytes": 1179275,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2755006959996535,
+      "bytes": 7429804,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -113,3 +113,4 @@
 {"step": 291500, "loss": 0.565649164840579, "lr": 0.0003, "grad_norm": 0.1470336876474914, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42667207381, "tokens_per_second": 150758.1432170335, "data_wait_s": 8.61400258145295e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 291600, "loss": 0.6152579374611378, "lr": 0.0003, "grad_norm": 0.18157318205932949, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42681906131, "tokens_per_second": 150530.93471218826, "data_wait_s": 9.913997928379104e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 291700, "loss": 0.5707413833588362, "lr": 0.0003, "grad_norm": 0.14450361581721619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42696551761, "tokens_per_second": 149938.74250766585, "data_wait_s": 1.3040000339969993e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 291800, "loss": 0.5925162993371487, "lr": 0.0003, "grad_norm": 0.15542837092778156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42711194406, "tokens_per_second": 150613.9168806891, "data_wait_s": 9.371997293783352e-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:d244c495019f99fef1e3ee25bd09a6bfa2d93e8b7ff7425210fd2f41127376c3
-size 245301
+oid sha256:d86e5e244a83239d983570d2fa122c22f0fbdfbfb3ca02dd1b45b85d8ad60287
+size 247419