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": 75964.15602325292,
-      "required_pages_per_second_upper_estimate": 0.006584380514702311
+      "required_token_positions_per_second": 75703.8137889036,
+      "required_pages_per_second_upper_estimate": 0.006561814709660273
     },
     {
       "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": 9495.519502906614,
-      "required_pages_per_second_upper_estimate": 0.0037879047003776187
+      "required_token_positions_per_second": 9462.97672361295,
+      "required_pages_per_second_upper_estimate": 0.0037749228991594665
     },
     {
       "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": 18991.03900581323,
-      "required_pages_per_second_upper_estimate": 0.00439324264312379
+      "required_token_positions_per_second": 18925.9534472259,
+      "required_pages_per_second_upper_estimate": 0.00437818624461132
     },
     {
       "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": 28486.55850871984,
-      "required_pages_per_second_upper_estimate": 0.0017346819557592212
+      "required_token_positions_per_second": 28388.93017083885,
+      "required_pages_per_second_upper_estimate": 0.0017287369021985715
     },
     {
       "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": 28486.55850871984,
-      "required_pages_per_second_upper_estimate": 0.0021929741298422037
+      "required_token_positions_per_second": 28388.93017083885,
+      "required_pages_per_second_upper_estimate": 0.002185458430139589
     },
     {
       "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": 28486.55850871984,
-      "required_pages_per_second_upper_estimate": 0.3537077182999099
+      "required_token_positions_per_second": 28388.93017083885,
+      "required_pages_per_second_upper_estimate": 0.3524955010844562
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.13243488976564777,
-    "producer_p95_seconds": 0.3283783129991207,
-    "gpu_update_seconds": 0.9705630110263002,
-    "producer_batches_per_second": 7.55088029876089,
-    "gpu_batches_per_second": 1.0303298071730267,
-    "producer_headroom": 7.328605118664539,
+    "producer_batch_seconds": 0.10452180185941984,
+    "producer_p95_seconds": 0.3406817700015381,
+    "gpu_update_seconds": 0.973900736435643,
+    "producer_batches_per_second": 9.567381945299642,
+    "gpu_batches_per_second": 1.0267986896281411,
+    "producer_headroom": 9.317680322288396,
     "sustainable": true,
-    "buffer_seconds": 60.17490668363061,
+    "buffer_seconds": 60.38184565900987,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 286918.1228622732,
+    "generation_bytes_per_second": 299548.84641915036,
     "upload_bytes_per_second": 3925264.591360211,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.56595005417964,
+    "estimated_full_shards_per_hour": 64.2762093012894,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -254,3 +254,4 @@
 {"step": 140400, "loss": 0.5918611697852612, "lr": 0.0003, "grad_norm": 0.14893381151320165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20547204317, "tokens_per_second": 147715.85919226793, "data_wait_s": 1.4041997928870842e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 140500, "loss": 0.5549229513853788, "lr": 0.0003, "grad_norm": 0.12578476302069969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20561851641, "tokens_per_second": 149664.62252536995, "data_wait_s": 1.4966997696319595e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 140600, "loss": 0.5978204291313887, "lr": 0.0003, "grad_norm": 0.13984901672717093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20576622327, "tokens_per_second": 150139.21753891415, "data_wait_s": 1.62090000230819e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 140700, "loss": 0.62378946878016, "lr": 0.0003, "grad_norm": 0.13608952735719315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20591427571, "tokens_per_second": 150778.95619423408, "data_wait_s": 1.3404001947492361e-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 sha256:13a4d163be27dec332fc5b738df7231e6070c1c786342776d9f2a7018568bf9d
-size 546288
+oid sha256:aadff28ea11e480eacbc188f806b6de3b20c69c9b386b5fc95f4786167de0380
+size 548406