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": 74035.15815188002,
-      "required_pages_per_second_upper_estimate": 0.006417179868212116
+      "required_token_positions_per_second": 74507.80339227885,
+      "required_pages_per_second_upper_estimate": 0.006458147559741476
     },
     {
       "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": 9254.394768985003,
-      "required_pages_per_second_upper_estimate": 0.003691716438880247
+      "required_token_positions_per_second": 9313.475424034856,
+      "required_pages_per_second_upper_estimate": 0.0037152845955141438
     },
     {
       "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": 18508.789537970006,
-      "required_pages_per_second_upper_estimate": 0.004281682716039017
+      "required_token_positions_per_second": 18626.95084806971,
+      "required_pages_per_second_upper_estimate": 0.004309017255562552
     },
     {
       "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": 27763.18430695501,
-      "required_pages_per_second_upper_estimate": 0.001690632262122867
+      "required_token_positions_per_second": 27940.426272104567,
+      "required_pages_per_second_upper_estimate": 0.0017014253678837518
     },
     {
       "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": 27763.18430695501,
-      "required_pages_per_second_upper_estimate": 0.0021372867813623953
+      "required_token_positions_per_second": 27940.426272104567,
+      "required_pages_per_second_upper_estimate": 0.0021509313584767704
     },
     {
       "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": 27763.18430695501,
-      "required_pages_per_second_upper_estimate": 0.344725831691707
+      "required_token_positions_per_second": 27940.426272104567,
+      "required_pages_per_second_upper_estimate": 0.34692658370816604
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.09300470426569518,
-    "producer_p95_seconds": 0.24965047900332138,
-    "gpu_update_seconds": 0.9958511853077979,
-    "producer_batches_per_second": 10.752144290929705,
-    "gpu_batches_per_second": 1.004166099065213,
-    "producer_headroom": 10.707535636722818,
+    "producer_batch_seconds": 0.11281403064811002,
+    "producer_p95_seconds": 0.27807193199987523,
+    "gpu_update_seconds": 0.98953393662442,
+    "producer_batches_per_second": 8.864145658612305,
+    "gpu_batches_per_second": 1.0105767604204487,
+    "producer_headroom": 8.771372948378897,
     "sustainable": true,
-    "buffer_seconds": 60.746922303775676,
+    "buffer_seconds": 60.36157013408962,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 393941.211286718,
+    "generation_bytes_per_second": 379190.8578050472,
     "upload_bytes_per_second": 5858887.1207046425,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.53061346007496,
+    "estimated_full_shards_per_hour": 81.36553097356378,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 374600, "loss": 0.6049759332090616, "lr": 0.0003, "grad_norm": 0.17264880688967682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54830715951, "tokens_per_second": 149078.2490447577, "data_wait_s": 1.0477000614628196e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 374700, "loss": 0.5894653387367725, "lr": 0.0003, "grad_norm": 0.1596533167155301, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54845284765, "tokens_per_second": 148263.97386253468, "data_wait_s": 1.0391002433607355e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 374800, "loss": 0.6016331817954779, "lr": 0.0003, "grad_norm": 0.15868112770444265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54859839341, "tokens_per_second": 148337.2599592984, "data_wait_s": 1.410699769621715e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 374900, "loss": 0.5981489419937134, "lr": 0.0003, "grad_norm": 0.14830377690576205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54874543805, "tokens_per_second": 150633.8736510711, "data_wait_s": 1.1036001524189487e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9470683d4dd5bbfb86380209c9ca97f366816db0af0bcd951520a65f231d6c51
-size 421878
+oid sha256:5cbb04974c9a3e1ddd1d2a50826ab8fd2b24580346b79b40803c9a128a88d871
+size 423996