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": 75963.5828920917,
-      "required_pages_per_second_upper_estimate": 0.006584330837145844
+      "required_token_positions_per_second": 75661.76059331131,
+      "required_pages_per_second_upper_estimate": 0.006558169645249183
     },
     {
       "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": 9495.447861511462,
-      "required_pages_per_second_upper_estimate": 0.0037878761215539582
+      "required_token_positions_per_second": 9457.720074163914,
+      "required_pages_per_second_upper_estimate": 0.0037728259431003327
     },
     {
       "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": 18990.895723022924,
-      "required_pages_per_second_upper_estimate": 0.004393209497171944
+      "required_token_positions_per_second": 18915.440148327827,
+      "required_pages_per_second_upper_estimate": 0.004375754177938992
     },
     {
       "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": 28486.343584534385,
-      "required_pages_per_second_upper_estimate": 0.001734668868003947
+      "required_token_positions_per_second": 28373.16022249174,
+      "required_pages_per_second_upper_estimate": 0.0017277765950827573
     },
     {
       "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": 28486.343584534385,
-      "required_pages_per_second_upper_estimate": 0.0021929575843834593
+      "required_token_positions_per_second": 28373.16022249174,
+      "required_pages_per_second_upper_estimate": 0.0021842444158618173
     },
     {
       "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": 28486.343584534385,
-      "required_pages_per_second_upper_estimate": 0.3537050496608315
+      "required_token_positions_per_second": 28373.16022249174,
+      "required_pages_per_second_upper_estimate": 0.35229969110460707
     }
   ],
   "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": 0.2019258627186531,
-    "producer_p95_seconds": 0.253017703000296,
-    "gpu_update_seconds": 0.9705703337444286,
-    "producer_batches_per_second": 4.952312628686489,
-    "gpu_batches_per_second": 1.030322033584143,
-    "producer_headroom": 4.806567720830994,
+    "producer_batch_seconds": 0.05907964213288608,
+    "producer_p95_seconds": 0.1968358600006468,
+    "gpu_update_seconds": 0.9744420354727741,
+    "producer_batches_per_second": 16.92630428855222,
+    "gpu_batches_per_second": 1.0262283066584108,
+    "producer_headroom": 16.49370240396837,
     "sustainable": true,
-    "buffer_seconds": 60.17536069215457,
+    "buffer_seconds": 60.415406199311995,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356951.8064929247,
-    "upload_bytes_per_second": 5670717.779836784,
+    "generation_bytes_per_second": 414929.6454920391,
+    "upload_bytes_per_second": 5674773.773016567,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.5935482606011,
+    "estimated_full_shards_per_hour": 89.03424285479431,
     "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": 0.999563274999673,
-      "bytes": 6224997,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.0198160460004146,
       "bytes": 6464312,
@@ -562,6 +556,12 @@
       "bytes": 7647995,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0442986829984875,
+      "bytes": 6780113,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -79,3 +79,4 @@
 {"step": 288100, "loss": 0.5865795081481338, "lr": 0.0003, "grad_norm": 0.15067789442631074, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42169651129, "tokens_per_second": 150945.08818430954, "data_wait_s": 1.127500036091078e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 288200, "loss": 0.6017746590077877, "lr": 0.0003, "grad_norm": 0.14384964663836383, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42184330783, "tokens_per_second": 150571.6507652425, "data_wait_s": 9.387998943566345e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 288300, "loss": 0.5978507325053215, "lr": 0.0003, "grad_norm": 0.13419214173368446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42198897066, "tokens_per_second": 148901.08860646313, "data_wait_s": 9.408999176230282e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 288400, "loss": 0.571783060207963, "lr": 0.0003, "grad_norm": 0.14598619365887963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42213362594, "tokens_per_second": 148148.08819389716, "data_wait_s": 1.1122998330392875e-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:bb4af816433c7fda192a47c53d4fecf4539ce81eac72fbc3871689a94c19c064
-size 172506
+oid sha256:77d516c43eca40a2751d174f3ce50f8d17baf5e0cd13bbbbe3189a2cd239d93c
+size 174624