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": 75230.93929262689,
-      "required_pages_per_second_upper_estimate": 0.006520827146812461
+      "required_token_positions_per_second": 75850.21824718612,
+      "required_pages_per_second_upper_estimate": 0.006574504677045494
     },
     {
       "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": 9403.867411578361,
-      "required_pages_per_second_upper_estimate": 0.0037513433108258983
+      "required_token_positions_per_second": 9481.277280898265,
+      "required_pages_per_second_upper_estimate": 0.003782223265078293
     },
     {
       "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": 18807.734823156723,
-      "required_pages_per_second_upper_estimate": 0.004350838446509639
+      "required_token_positions_per_second": 18962.55456179653,
+      "required_pages_per_second_upper_estimate": 0.004386653268309619
     },
     {
       "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": 28211.602234735084,
-      "required_pages_per_second_upper_estimate": 0.0017179385612576304
+      "required_token_positions_per_second": 28443.83184269479,
+      "required_pages_per_second_upper_estimate": 0.0017320801259677876
     },
     {
       "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": 28211.602234735084,
-      "required_pages_per_second_upper_estimate": 0.0021718072347431767
+      "required_token_positions_per_second": 28443.83184269479,
+      "required_pages_per_second_upper_estimate": 0.0021896849128166143
     },
     {
       "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": 28211.602234735084,
-      "required_pages_per_second_upper_estimate": 0.35029368159647223
+      "required_token_positions_per_second": 28443.83184269479,
+      "required_pages_per_second_upper_estimate": 0.3531771961048312
     }
   ],
   "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": 0.10856471481284302,
-    "producer_p95_seconds": 0.36645041900192155,
-    "gpu_update_seconds": 0.9800223245016138,
-    "producer_batches_per_second": 9.2110959046309,
-    "gpu_batches_per_second": 1.020384918791055,
-    "producer_headroom": 9.02707961966367,
+    "producer_batch_seconds": 0.33411695057796464,
+    "producer_p95_seconds": 0.33604296399789746,
+    "gpu_update_seconds": 0.9720209342012692,
+    "producer_batches_per_second": 2.9929639854253804,
+    "gpu_batches_per_second": 1.028784427180801,
+    "producer_headroom": 2.9092236491439323,
     "sustainable": true,
-    "buffer_seconds": 60.76138411910005,
+    "buffer_seconds": 60.265297920478694,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "upload_pages_per_commit": 159,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 669071.9587169519,
+    "generation_bytes_per_second": 614223.996231387,
     "upload_bytes_per_second": 5137223.934947517,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.78363357129773,
+    "estimated_full_shards_per_hour": 65.89908559420685,
     "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

@@ -304,3 +304,4 @@
 {"step": 145400, "loss": 0.6292114295065403, "lr": 0.0003, "grad_norm": 0.1463837511759007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21279195490, "tokens_per_second": 151062.97558532076, "data_wait_s": 1.0263000149279833e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 145500, "loss": 0.574711250141263, "lr": 0.0003, "grad_norm": 0.16296621779060388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21293792544, "tokens_per_second": 148855.49993111668, "data_wait_s": 9.462997695663944e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 145600, "loss": 0.6238197414204478, "lr": 0.0003, "grad_norm": 0.15038211799517512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21308343245, "tokens_per_second": 149403.89361964725, "data_wait_s": 1.0685998859116808e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 145700, "loss": 0.5823377333581448, "lr": 0.0003, "grad_norm": 0.14650722565116236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21322804932, "tokens_per_second": 147950.65339411545, "data_wait_s": 1.0825999197550118e-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 sha256:2795db37562a889bc1a6a90beb453367501f0aa9da78919891c0970e7a33d3a9
-size 652971
+oid sha256:5569377f97c434ec73f760eca322a4df48dbfe220882f36015f389769632f035
+size 655089