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": 74665.7184177298,
-      "required_pages_per_second_upper_estimate": 0.006471835233915592
+      "required_token_positions_per_second": 74727.6826879381,
+      "required_pages_per_second_upper_estimate": 0.00647720613981024
     },
     {
       "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": 9333.214802216226,
-      "required_pages_per_second_upper_estimate": 0.003723158928600696
+      "required_token_positions_per_second": 9340.960335992262,
+      "required_pages_per_second_upper_estimate": 0.0037262487378300073
     },
     {
       "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": 18666.42960443245,
-      "required_pages_per_second_upper_estimate": 0.00431814997104469
+      "required_token_positions_per_second": 18681.920671984524,
+      "required_pages_per_second_upper_estimate": 0.004321733556889386
     },
     {
       "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": 27999.64440664868,
-      "required_pages_per_second_upper_estimate": 0.001705031441583948
+      "required_token_positions_per_second": 28022.881007976786,
+      "required_pages_per_second_upper_estimate": 0.0017064464286918044
     },
     {
       "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": 27999.64440664868,
-      "required_pages_per_second_upper_estimate": 0.0021554901344002643
+      "required_token_positions_per_second": 28022.881007976786,
+      "required_pages_per_second_upper_estimate": 0.0021572789522935312
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27999.64440664868,
-      "required_pages_per_second_upper_estimate": 0.3476618747488568
+      "required_token_positions_per_second": 28022.881007976786,
+      "required_pages_per_second_upper_estimate": 0.3479503955694499
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11119139293735714,
-    "producer_p95_seconds": 0.21749101299974427,
-    "gpu_update_seconds": 0.9874411116962193,
-    "producer_batches_per_second": 8.993501867211778,
-    "gpu_batches_per_second": 1.0127186200321425,
-    "producer_headroom": 8.880553481801623,
+    "producer_batch_seconds": 0.22971054428913362,
+    "producer_p95_seconds": 0.2499871380005061,
+    "gpu_update_seconds": 0.9866223245258018,
+    "producer_batches_per_second": 4.353304734419649,
+    "gpu_batches_per_second": 1.0135590642352714,
+    "producer_headroom": 4.295067636442293,
     "sustainable": true,
-    "buffer_seconds": 60.23390781346938,
+    "buffer_seconds": 60.18396179607391,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 427927.76767154766,
-    "upload_bytes_per_second": 5970001.893056868,
+    "generation_bytes_per_second": 414612.6208023374,
+    "upload_bytes_per_second": 5994103.733507505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.82333729371855,
+    "estimated_full_shards_per_hour": 88.9662167363414,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 11636309,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0268402850015264,
+      "bytes": 7548268,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -111,3 +111,4 @@
 {"step": 126100, "loss": 0.5676320344209671, "lr": 0.0003, "grad_norm": 0.12993174176535047, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18454002821, "tokens_per_second": 147875.12192090612, "data_wait_s": 1.135099955718033e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126200, "loss": 0.5653232540935278, "lr": 0.0003, "grad_norm": 0.1336714102744638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18468457288, "tokens_per_second": 147792.7347952092, "data_wait_s": 1.3756998669123277e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 126300, "loss": 0.5899433512240648, "lr": 0.0003, "grad_norm": 0.1420256776911786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18483034000, "tokens_per_second": 147523.62749179584, "data_wait_s": 1.979500120796729e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 126400, "loss": 0.5922726560384035, "lr": 0.0003, "grad_norm": 0.15503752254612066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18497771948, "tokens_per_second": 150291.41941792518, "data_wait_s": 1.3797000065096654e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3353189269a105731b41d95a8ac2acf185d5566880f155e73c98c4ac1f7492b7
-size 241065
+oid sha256:bd321af1fe78224b2e36032d57264056c5acdc0cf8929fe0a784a71bebd7f537
+size 243183