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": 75910.77989921649,
-      "required_pages_per_second_upper_estimate": 0.006579754007551383
+      "required_token_positions_per_second": 75862.18750229837,
+      "required_pages_per_second_upper_estimate": 0.006575542141742825
     },
     {
       "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": 9488.847487402061,
-      "required_pages_per_second_upper_estimate": 0.003785243133637331
+      "required_token_positions_per_second": 9482.773437787297,
+      "required_pages_per_second_upper_estimate": 0.003782820104430867
     },
     {
       "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": 18977.694974804122,
-      "required_pages_per_second_upper_estimate": 0.0043901557363967576
+      "required_token_positions_per_second": 18965.546875574593,
+      "required_pages_per_second_upper_estimate": 0.00438734548743925
     },
     {
       "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": 28466.542462206187,
-      "required_pages_per_second_upper_estimate": 0.0017334630835426196
+      "required_token_positions_per_second": 28448.320313361888,
+      "required_pages_per_second_upper_estimate": 0.0017323534502822282
     },
     {
       "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": 28466.542462206187,
-      "required_pages_per_second_upper_estimate": 0.0021914332391736197
+      "required_token_positions_per_second": 28448.320313361888,
+      "required_pages_per_second_upper_estimate": 0.002190030447713449
     },
     {
       "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": 28466.542462206187,
-      "required_pages_per_second_upper_estimate": 0.3534591859916087
+      "required_token_positions_per_second": 28448.320313361888,
+      "required_pages_per_second_upper_estimate": 0.3532329278885716
     }
   ],
   "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": 0.06820318753139532,
-    "producer_p95_seconds": 0.2179764870015788,
-    "gpu_update_seconds": 0.9712454554924284,
-    "producer_batches_per_second": 14.662071322395008,
-    "gpu_batches_per_second": 1.0296058471573417,
-    "producer_headroom": 14.24047013998201,
+    "producer_batch_seconds": 0.11627025199993568,
+    "producer_p95_seconds": 0.26189117499961867,
+    "gpu_update_seconds": 0.9718675723365647,
+    "producer_batches_per_second": 8.600652211543785,
+    "gpu_batches_per_second": 1.0289467705932396,
+    "producer_headroom": 8.358694985344165,
     "sustainable": true,
-    "buffer_seconds": 60.217218240530556,
+    "buffer_seconds": 60.255789484867016,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370270.8891893889,
+    "generation_bytes_per_second": 385701.97946807713,
     "upload_bytes_per_second": 3946120.9874398075,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.45151335488558,
+    "estimated_full_shards_per_hour": 82.76266611129509,
     "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

@@ -128,3 +128,4 @@
 {"step": 293000, "loss": 0.5930706281214952, "lr": 0.0003, "grad_norm": 0.1754426991099345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42886873471, "tokens_per_second": 151069.63870371747, "data_wait_s": 9.398998372489586e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 293100, "loss": 0.5831968560814857, "lr": 0.0003, "grad_norm": 0.15030196289962997, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42901558737, "tokens_per_second": 150688.91296613295, "data_wait_s": 1.6109999705804512e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 293200, "loss": 0.5806980840861797, "lr": 0.0003, "grad_norm": 0.14956070229021243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42916220789, "tokens_per_second": 150175.45408458274, "data_wait_s": 1.2565000361064449e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 293300, "loss": 0.5915417168289423, "lr": 0.0003, "grad_norm": 0.1528488919679106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42930985270, "tokens_per_second": 152309.3695231858, "data_wait_s": 1.55180023284629e-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 sha256:54e949807dbf6c1f96076aa3ca758c2e7ec285bd6e33f51f43d350db9fb9bc80
-size 277071
+oid sha256:e49e61a87963eeb25284f052f104faff13ed59cad604d9e727ff5bbddbd4c123
+size 279189