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": 77052.93697440557,
-      "required_pages_per_second_upper_estimate": 0.005330700967335261
+      "required_token_positions_per_second": 76295.78030360074,
+      "required_pages_per_second_upper_estimate": 0.00527831911200346
     },
     {
       "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": 9631.617121800697,
-      "required_pages_per_second_upper_estimate": 0.004032556957861076
+      "required_token_positions_per_second": 9536.972537950092,
+      "required_pages_per_second_upper_estimate": 0.003992931247006483
     },
     {
       "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": 19263.234243601393,
-      "required_pages_per_second_upper_estimate": 0.0040120236872932665
+      "required_token_positions_per_second": 19073.945075900185,
+      "required_pages_per_second_upper_estimate": 0.003972599745552147
     },
     {
       "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": 28894.85136540209,
-      "required_pages_per_second_upper_estimate": 0.0017322022194724445
+      "required_token_positions_per_second": 28610.917613850273,
+      "required_pages_per_second_upper_estimate": 0.0017151808246086487
     },
     {
       "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": 28894.85136540209,
-      "required_pages_per_second_upper_estimate": 0.0022464475986406717
+      "required_token_positions_per_second": 28610.917613850273,
+      "required_pages_per_second_upper_estimate": 0.002224372998350104
     },
     {
       "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": 28894.85136540209,
-      "required_pages_per_second_upper_estimate": 0.37063212844117044
+      "required_token_positions_per_second": 28610.917613850273,
+      "required_pages_per_second_upper_estimate": 0.36699013114057377
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05528957291409142,
-    "producer_p95_seconds": 0.17510776000017358,
-    "gpu_update_seconds": 0.9568486665795749,
-    "producer_batches_per_second": 18.08659295585071,
-    "gpu_batches_per_second": 1.0450973439453881,
-    "producer_headroom": 17.306132352773282,
+    "producer_batch_seconds": 0.38075650120313753,
+    "producer_p95_seconds": 0.17078465200029314,
+    "gpu_update_seconds": 0.9663443994755297,
+    "producer_batches_per_second": 2.6263504282662,
+    "gpu_batches_per_second": 1.034827749343543,
+    "producer_headroom": 2.537959027415201,
     "sustainable": true,
-    "buffer_seconds": 60.28146599451322,
+    "buffer_seconds": 60.87969716695837,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 447280.0853415409,
+    "generation_bytes_per_second": 391819.6533964042,
     "upload_bytes_per_second": 5833319.346352694,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.97589416680022,
+    "estimated_full_shards_per_hour": 84.07537652415365,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -47,3 +47,4 @@
 {"step": 489700, "loss": 0.5630510449409485, "lr": 0.0003, "grad_norm": 0.142588708954507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71675674359, "tokens_per_second": 153290.3542468922, "data_wait_s": 1.1117000212834682e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 489800, "loss": 0.5368896126747131, "lr": 0.0003, "grad_norm": 0.1608872092958825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71690391291, "tokens_per_second": 153667.2404844987, "data_wait_s": 1.6226999832724687e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 489900, "loss": 0.6282123327255249, "lr": 0.0003, "grad_norm": 0.17659918603157956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71704929021, "tokens_per_second": 144403.0062497457, "data_wait_s": 8.087999958661385e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 490000, "loss": 0.6224940419197083, "lr": 0.0003, "grad_norm": 0.16324780468432967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71719430326, "tokens_per_second": 151494.72346230017, "data_wait_s": 7.556000127806328e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6927f405e1b1a5f6a40d23c64a203d78ca48c4bc78b52061a2bae4c487f1186c
-size 103947
+oid sha256:c6f495b7c9409d4b43f0b54d341cf1efefbbbf293bb66a3002768e8acd9436cd
+size 106065