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": 75700.44897128262,
-      "required_pages_per_second_upper_estimate": 0.006561523055796939
+      "required_token_positions_per_second": 75842.61374065137,
+      "required_pages_per_second_upper_estimate": 0.006573845537692509
     },
     {
       "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": 9462.556121410327,
-      "required_pages_per_second_upper_estimate": 0.0037747551146522767
+      "required_token_positions_per_second": 9480.326717581422,
+      "required_pages_per_second_upper_estimate": 0.003781844071159016
     },
     {
       "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": 18925.112242820655,
-      "required_pages_per_second_upper_estimate": 0.004377991646776858
+      "required_token_positions_per_second": 18960.653435162843,
+      "required_pages_per_second_upper_estimate": 0.004386213476121602
     },
     {
       "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": 28387.66836423098,
-      "required_pages_per_second_upper_estimate": 0.0017286600647963399
+      "required_token_positions_per_second": 28440.98015274427,
+      "required_pages_per_second_upper_estimate": 0.0017319064730114617
     },
     {
       "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": 28387.66836423098,
-      "required_pages_per_second_upper_estimate": 0.002185361292774011
+      "required_token_positions_per_second": 28440.98015274427,
+      "required_pages_per_second_upper_estimate": 0.00218946538183024
     },
     {
       "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": 28387.66836423098,
-      "required_pages_per_second_upper_estimate": 0.3524798336693815
+      "required_token_positions_per_second": 28440.98015274427,
+      "required_pages_per_second_upper_estimate": 0.35314178765963805
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06993858995321034,
-    "producer_p95_seconds": 0.3000901539999177,
-    "gpu_update_seconds": 0.9739440254570898,
-    "producer_batches_per_second": 14.298257952712667,
-    "gpu_batches_per_second": 1.026753051368308,
-    "producer_headroom": 13.925702907488823,
-    "sustainable": true,
-    "buffer_seconds": 60.38452957833957,
-    "buffer_drain_seconds": null,
+    "producer_batch_seconds": 1.1080608083517234,
+    "producer_p95_seconds": 0.32552214799943613,
+    "gpu_update_seconds": 0.9721183957625401,
+    "producer_batches_per_second": 0.9024775467760949,
+    "gpu_batches_per_second": 1.0286812844597897,
+    "producer_headroom": 0.8773150249836902,
+    "sustainable": false,
+    "buffer_seconds": 60.27134053727749,
+    "buffer_drain_seconds": 491.2691267147016,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269774.2521448895,
-    "upload_bytes_per_second": 2684193.9707094296,
+    "generation_bytes_per_second": 219705.554414035,
+    "upload_bytes_per_second": 2534349.6634966815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.887274487114084,
+    "estimated_full_shards_per_hour": 47.14369749370372,
     "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": 1.1376306590027525,
-      "bytes": 968236,
-      "pages": 28,
-      "shards": 1
-    },
-    {
-      "seconds": 1.3645541389996652,
-      "bytes": 7442538,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.78800499800127,
       "bytes": 976266,
@@ -562,6 +550,18 @@
       "bytes": 6143251,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 6.752744052999333,
+      "bytes": 5959933,
+      "pages": 27,
+      "shards": 2
+    },
+    {
+      "seconds": 8.626221913000336,
+      "bytes": 11949436,
+      "pages": 25,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -161,3 +161,4 @@
 {"step": 296300, "loss": 0.6124036759138107, "lr": 0.0003, "grad_norm": 0.17343848056638297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43370467936, "tokens_per_second": 149906.69870673746, "data_wait_s": 1.152500044554472e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 296400, "loss": 0.608258318156004, "lr": 0.0003, "grad_norm": 0.14973363631122263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43385143040, "tokens_per_second": 150896.99945533354, "data_wait_s": 1.6546000551898032e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 296500, "loss": 0.6336312182247639, "lr": 0.0003, "grad_norm": 0.15525307091539312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43399781144, "tokens_per_second": 150483.4825429867, "data_wait_s": 1.135300044552423e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 296600, "loss": 0.6367071159183979, "lr": 0.0003, "grad_norm": 0.1494025677632923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43414573517, "tokens_per_second": 80757.99840457646, "data_wait_s": 1.1347001418471336e-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 sha256:618545f9b703357990e076633902050f95e9472a1d2ed92a04c62e4c225fe9cb
-size 347748
+oid sha256:5c72ce002bc9b309bbbcd2a51ef54318af94c36bff09eadc76570862ea81715c
+size 349866