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": 74631.48414961596,
-      "required_pages_per_second_upper_estimate": 0.006468867894321444
+      "required_token_positions_per_second": 74640.9393182521,
+      "required_pages_per_second_upper_estimate": 0.006469687444374919
     },
     {
       "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": 9328.935518701996,
-      "required_pages_per_second_upper_estimate": 0.0037214518584258796
+      "required_token_positions_per_second": 9330.117414781513,
+      "required_pages_per_second_upper_estimate": 0.0037219233344429202
     },
     {
       "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": 18657.87103740399,
-      "required_pages_per_second_upper_estimate": 0.0043161700971882915
+      "required_token_positions_per_second": 18660.234829563025,
+      "required_pages_per_second_upper_estimate": 0.004316716918902963
     },
     {
       "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": 27986.806556105985,
-      "required_pages_per_second_upper_estimate": 0.0017042496838408942
+      "required_token_positions_per_second": 27990.35224434454,
+      "required_pages_per_second_upper_estimate": 0.0017044655976518337
     },
     {
       "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": 27986.806556105985,
-      "required_pages_per_second_upper_estimate": 0.0021545018411280427
+      "required_token_positions_per_second": 27990.35224434454,
+      "required_pages_per_second_upper_estimate": 0.00215477479802374
     },
     {
       "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": 27986.806556105985,
-      "required_pages_per_second_upper_estimate": 0.3475024716106384
+      "required_token_positions_per_second": 27990.35224434454,
+      "required_pages_per_second_upper_estimate": 0.34754649719190606
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11873437012479826,
-    "producer_p95_seconds": 0.27226086299924646,
-    "gpu_update_seconds": 0.987894061602678,
-    "producer_batches_per_second": 8.422161156444666,
-    "gpu_batches_per_second": 1.0122542880536018,
-    "producer_headroom": 8.32020299231243,
+    "producer_batch_seconds": 0.0947450690233893,
+    "producer_p95_seconds": 0.30543813499753014,
+    "gpu_update_seconds": 0.9877689197564953,
+    "producer_batches_per_second": 10.554638993963204,
+    "gpu_batches_per_second": 1.0123825319858413,
+    "producer_headroom": 10.425544357486816,
     "sustainable": true,
-    "buffer_seconds": 60.26153775776336,
+    "buffer_seconds": 60.25390410514621,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348065.5543013182,
-    "upload_bytes_per_second": 5429514.796567988,
+    "generation_bytes_per_second": 395049.05262344395,
+    "upload_bytes_per_second": 5490991.8692655815,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.68676540164623,
+    "estimated_full_shards_per_hour": 84.7683304217099,
     "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": 0.9418184709993511,
-      "bytes": 1092329,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.2206816289999551,
       "bytes": 9827767,
@@ -562,6 +556,12 @@
       "bytes": 2378485,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.0251769999995304,
+      "bytes": 6275188,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -207,3 +207,4 @@
 {"step": 257800, "loss": 0.5910930745303631, "lr": 0.0003, "grad_norm": 0.16328892842084045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37734635326, "tokens_per_second": 149862.3811322859, "data_wait_s": 1.4048000593902543e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 257900, "loss": 0.5136015452444553, "lr": 0.0003, "grad_norm": 0.19319863370097987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37749359212, "tokens_per_second": 149930.80414826045, "data_wait_s": 1.42680000863038e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 258000, "loss": 0.585047347471118, "lr": 0.0003, "grad_norm": 0.15418200421276249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37764114907, "tokens_per_second": 150328.98359405634, "data_wait_s": 1.2985998182557523e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 258100, "loss": 0.6123698111623526, "lr": 0.0003, "grad_norm": 0.15422787591436724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37778839572, "tokens_per_second": 148994.77597158717, "data_wait_s": 1.5338999219238758e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6beefdb9e2efb1bc8edbe74c995f893611af2298d1f8aef01a606f7b9c9d61e5
-size 445959
+oid sha256:b598d10c5f62bbef60408356d69999be1b727e97b1867378c1f0838b39007825
+size 448077