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": 75306.65991166854,
-      "required_pages_per_second_upper_estimate": 0.0065273904181588375
+      "required_token_positions_per_second": 75684.4765577787,
+      "required_pages_per_second_upper_estimate": 0.006560138607476131
     },
     {
       "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": 9413.332488958567,
-      "required_pages_per_second_upper_estimate": 0.00375511907170838
+      "required_token_positions_per_second": 9460.559569722338,
+      "required_pages_per_second_upper_estimate": 0.00377395866033283
     },
     {
       "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": 18826.664977917135,
-      "required_pages_per_second_upper_estimate": 0.004355217604654116
+      "required_token_positions_per_second": 18921.119139444676,
+      "required_pages_per_second_upper_estimate": 0.004377067912587045
     },
     {
       "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": 28239.997466875706,
-      "required_pages_per_second_upper_estimate": 0.001719667681916725
+      "required_token_positions_per_second": 28381.678709167012,
+      "required_pages_per_second_upper_estimate": 0.0017282953262282357
     },
     {
       "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": 28239.997466875706,
-      "required_pages_per_second_upper_estimate": 0.002173993178316395
+      "required_token_positions_per_second": 28381.678709167012,
+      "required_pages_per_second_upper_estimate": 0.0021849001925467603
     },
     {
       "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": 28239.997466875706,
-      "required_pages_per_second_upper_estimate": 0.35064625534693006
+      "required_token_positions_per_second": 28381.678709167012,
+      "required_pages_per_second_upper_estimate": 0.352405462199573
     }
   ],
   "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": 0.08708365628899628,
-    "producer_p95_seconds": 0.2824931139984983,
-    "gpu_update_seconds": 0.9790369150149503,
-    "producer_batches_per_second": 11.4832110020897,
-    "gpu_batches_per_second": 1.0214119454165114,
-    "producer_headroom": 11.242487473951636,
+    "producer_batch_seconds": 0.08632017542160497,
+    "producer_p95_seconds": 0.3057781249990512,
+    "gpu_update_seconds": 0.9741495661097016,
+    "producer_batches_per_second": 11.584777198561058,
+    "gpu_batches_per_second": 1.026536411645219,
+    "producer_headroom": 11.285305681455819,
     "sustainable": true,
-    "buffer_seconds": 60.700288730926914,
+    "buffer_seconds": 60.3972730988015,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342980.6316557939,
-    "upload_bytes_per_second": 4013925.790408049,
+    "generation_bytes_per_second": 379466.7252959816,
+    "upload_bytes_per_second": 4006282.667876716,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.5956593728577,
+    "estimated_full_shards_per_hour": 81.4247257152518,
     "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": 1.218386226999428,
-      "bytes": 7220161,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 0.9670737249998638,
       "bytes": 5661886,
@@ -562,6 +556,12 @@
       "bytes": 2547314,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.1576964999985648,
+      "bytes": 6153703,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -134,3 +134,4 @@
 {"step": 293600, "loss": 0.6076928023248911, "lr": 0.0003, "grad_norm": 0.1773359722816775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42975015545, "tokens_per_second": 132135.99491782277, "data_wait_s": 1.7972000932786614e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 293700, "loss": 0.6051802970468998, "lr": 0.0003, "grad_norm": 0.16428482028590638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42989557176, "tokens_per_second": 149093.7518796874, "data_wait_s": 1.0316998668713495e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 293800, "loss": 0.5998360477387905, "lr": 0.0003, "grad_norm": 0.14458538680261215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43004171777, "tokens_per_second": 149164.57439300106, "data_wait_s": 8.318998879985884e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 293900, "loss": 0.6085012704133987, "lr": 0.0003, "grad_norm": 0.14893825711803013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43018855944, "tokens_per_second": 150892.0930995991, "data_wait_s": 1.2694999895757064e-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 sha256:6d2a207874b7cdf12a92d5cdf19e33136d133adb07472f7f1705157f4462cf45
-size 289779
+oid sha256:b6d0c098a0c5e431e12d990ccdf12442f37b3a0877d05990717003d040d6e724
+size 291897