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": 74888.88786573302,
-      "required_pages_per_second_upper_estimate": 0.006491178996050716
+      "required_token_positions_per_second": 75191.36232221208,
+      "required_pages_per_second_upper_estimate": 0.00651739671532382
     },
     {
       "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": 9361.110983216628,
-      "required_pages_per_second_upper_estimate": 0.003734287132286831
+      "required_token_positions_per_second": 9398.92029027651,
+      "required_pages_per_second_upper_estimate": 0.0037493698301725347
     },
     {
       "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": 18722.221966433255,
-      "required_pages_per_second_upper_estimate": 0.004331056552081542
+      "required_token_positions_per_second": 18797.84058055302,
+      "required_pages_per_second_upper_estimate": 0.004348549587615995
     },
     {
       "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": 28083.332949649885,
-      "required_pages_per_second_upper_estimate": 0.0017101276347728702
+      "required_token_positions_per_second": 28196.76087082953,
+      "required_pages_per_second_upper_estimate": 0.0017170348000624008
     },
     {
       "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": 28083.332949649885,
-      "required_pages_per_second_upper_estimate": 0.0021619327101052113
+      "required_token_positions_per_second": 28196.76087082953,
+      "required_pages_per_second_upper_estimate": 0.0021706647054662025
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28083.332949649885,
-      "required_pages_per_second_upper_estimate": 0.34870100636539586
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28196.76087082953,
+      "required_pages_per_second_upper_estimate": 0.350109401527615
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.15362248828921565,
-    "producer_p95_seconds": 0.37231110100401565,
-    "gpu_update_seconds": 0.9844985297710074,
-    "producer_batches_per_second": 6.509463628250581,
-    "gpu_batches_per_second": 1.0157455493941654,
-    "producer_headroom": 6.408557371610543,
+    "producer_batch_seconds": 0.5195615938906144,
+    "producer_p95_seconds": 0.36418312799651176,
+    "gpu_update_seconds": 0.980538159211144,
+    "producer_batches_per_second": 1.924699615519569,
+    "gpu_batches_per_second": 1.019848121774795,
+    "producer_headroom": 1.8872414180359547,
     "sustainable": true,
-    "buffer_seconds": 60.05441031603145,
+    "buffer_seconds": 60.79336587109093,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 325213.5955592776,
-    "upload_bytes_per_second": 5206538.460851879,
+    "generation_bytes_per_second": 288239.62925976276,
+    "upload_bytes_per_second": 5176381.674066845,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.78326702197786,
+    "estimated_full_shards_per_hour": 61.84951456398642,
     "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": 1.0368577580011333,
-      "bytes": 6834427,
-      "pages": 31,
-      "shards": 2
-    },
     {
       "seconds": 1.4560243180021644,
       "bytes": 7277270,
@@ -562,6 +556,12 @@
       "bytes": 13398838,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.170311688001675,
+      "bytes": 5100299,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -398,3 +398,4 @@
 {"step": 276900, "loss": 0.6357842348515987, "lr": 0.0003, "grad_norm": 0.1444372361515998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40530044420, "tokens_per_second": 150484.12332854012, "data_wait_s": 1.3664997823070735e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277000, "loss": 0.607653422281146, "lr": 0.0003, "grad_norm": 0.14279326064731315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40544820615, "tokens_per_second": 150069.12731706758, "data_wait_s": 1.3805998605675995e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277100, "loss": 0.5619482211768627, "lr": 0.0003, "grad_norm": 0.13645759299485968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40559593939, "tokens_per_second": 150571.18316649858, "data_wait_s": 1.503700332250446e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 277200, "loss": 0.672737043350935, "lr": 0.0003, "grad_norm": 0.19240088435973923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40574057787, "tokens_per_second": 123632.47096469428, "data_wait_s": 1.122899993788451e-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 sha256:315612afa8d935a360a7457e32a404fc7d64d03f796aec41ceea7c76db923afe
-size 853629
+oid sha256:405bd0c404200e5d76eabd40d02d4b38b338850a2ca93c90926b4588adcee0dc
+size 855747