Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76188.8127163753,
-      "required_pages_per_second_upper_estimate": 0.005270918846120206
+      "required_token_positions_per_second": 77384.02832590906,
+      "required_pages_per_second_upper_estimate": 0.0053536066352700504
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9523.601589546912,
-      "required_pages_per_second_upper_estimate": 0.003987333110127225
+      "required_token_positions_per_second": 9673.003540738633,
+      "required_pages_per_second_upper_estimate": 0.004049884587223686
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19047.203179093824,
-      "required_pages_per_second_upper_estimate": 0.003967030113678625
+      "required_token_positions_per_second": 19346.007081477266,
+      "required_pages_per_second_upper_estimate": 0.004029263086556285
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28570.804768640737,
-      "required_pages_per_second_upper_estimate": 0.0017127761207871052
+      "required_token_positions_per_second": 29019.0106222159,
+      "required_pages_per_second_upper_estimate": 0.0017396453773382244
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28570.804768640737,
-      "required_pages_per_second_upper_estimate": 0.0022212544010728223
+      "required_token_positions_per_second": 29019.0106222159,
+      "required_pages_per_second_upper_estimate": 0.0022561004347390907
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28570.804768640737,
-      "required_pages_per_second_upper_estimate": 0.3664756066320434
+      "required_token_positions_per_second": 29019.0106222159,
+      "required_pages_per_second_upper_estimate": 0.3722247100757546
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07892110987486944,
-    "producer_p95_seconds": 0.29066098499970394,
-    "gpu_update_seconds": 0.9677011279131484,
-    "producer_batches_per_second": 12.670881106278339,
-    "gpu_batches_per_second": 1.0333769085879896,
-    "producer_headroom": 12.26162593819895,
+    "producer_batch_seconds": 0.13244694707796612,
+    "producer_p95_seconds": 0.29343187599806697,
+    "gpu_update_seconds": 0.9527547427421146,
+    "producer_batches_per_second": 7.5501929041168525,
+    "gpu_batches_per_second": 1.0495880578058412,
+    "producer_headroom": 7.19348209801519,
     "sustainable": true,
-    "buffer_seconds": 60.965171058528355,
+    "buffer_seconds": 60.02354879275322,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 479650.96684436634,
-    "upload_bytes_per_second": 5143495.388498919,
+    "generation_bytes_per_second": 476524.12971234287,
+    "upload_bytes_per_second": 5110434.243277047,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.92193178175204,
+    "estimated_full_shards_per_hour": 102.2509853222629,
     "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": 2.1376686900002824,
-      "bytes": 11886724,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1407382979996328,
       "bytes": 5980776,
@@ -562,6 +556,12 @@
       "bytes": 6064094,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 0.8354571769996255,
+      "bytes": 2427677,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 466400, "loss": 0.5581997632980347, "lr": 0.0003, "grad_norm": 0.15003424692333678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68267135738, "tokens_per_second": 152192.67220994682, "data_wait_s": 1.2128999514970928e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466500, "loss": 0.5825616717338562, "lr": 0.0003, "grad_norm": 0.16656615871077907, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68281815191, "tokens_per_second": 152342.5657942288, "data_wait_s": 2.007999864872545e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466600, "loss": 0.5827690958976746, "lr": 0.0003, "grad_norm": 0.17281665825381523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68296585906, "tokens_per_second": 153156.93834537195, "data_wait_s": 1.5769001038279384e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 466700, "loss": 0.5531118512153625, "lr": 0.0003, "grad_norm": 0.14978028778130684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68311272192, "tokens_per_second": 152311.2333625329, "data_wait_s": 1.5227000403683633e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:68ac64608b26dfda9db939f537afce5c8a40a2564e69a1f748d4d9285b88e2c6
-size 563232
+oid sha256:86e2301d40e09ef72a8f06c2e566d5fcfa748e002df13f5840a83a9153a6f231
+size 565350