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": 75712.73263384549,
-      "required_pages_per_second_upper_estimate": 0.00656258777253523
+      "required_token_positions_per_second": 75648.76379971155,
+      "required_pages_per_second_upper_estimate": 0.006557043116120028
     },
     {
       "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": 9464.091579230686,
-      "required_pages_per_second_upper_estimate": 0.003775367631733958
+      "required_token_positions_per_second": 9456.095474963944,
+      "required_pages_per_second_upper_estimate": 0.0037721778661895422
     },
     {
       "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": 18928.183158461372,
-      "required_pages_per_second_upper_estimate": 0.004378702049064069
+      "required_token_positions_per_second": 18912.190949927888,
+      "required_pages_per_second_upper_estimate": 0.004375002533073105
     },
     {
       "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": 28392.274737692056,
-      "required_pages_per_second_upper_estimate": 0.0017289405687723456
+      "required_token_positions_per_second": 28368.28642489183,
+      "required_pages_per_second_upper_estimate": 0.0017274798063797625
     },
     {
       "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": 28392.274737692056,
-      "required_pages_per_second_upper_estimate": 0.0021857159041543026
+      "required_token_positions_per_second": 28368.28642489183,
+      "required_pages_per_second_upper_estimate": 0.0021838692174310412
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28392.274737692056,
-      "required_pages_per_second_upper_estimate": 0.35253702941122783
+      "required_token_positions_per_second": 28368.28642489183,
+      "required_pages_per_second_upper_estimate": 0.3522391748499675
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.13690956783588604,
-    "producer_p95_seconds": 0.3313065820002521,
-    "gpu_update_seconds": 0.9737860124076639,
-    "producer_batches_per_second": 7.304091421855216,
-    "gpu_batches_per_second": 1.0269196592047185,
-    "producer_headroom": 7.112622059949415,
+    "producer_batch_seconds": 0.0988835347266388,
+    "producer_p95_seconds": 0.329583447000914,
+    "gpu_update_seconds": 0.9746094489422592,
+    "producer_batches_per_second": 10.112907095852474,
+    "gpu_batches_per_second": 1.026052026363275,
+    "producer_headroom": 9.856134811893043,
     "sustainable": true,
-    "buffer_seconds": 60.37473276927516,
+    "buffer_seconds": 60.42578583442007,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284112.274000039,
-    "upload_bytes_per_second": 4763971.548796716,
+    "generation_bytes_per_second": 292015.31808550213,
+    "upload_bytes_per_second": 4767883.0582335405,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.96388020516278,
+    "estimated_full_shards_per_hour": 62.6596894924526,
     "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": 1.2147977610002272,
-      "bytes": 6193785,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 0.9691767799995432,
       "bytes": 2430413,
@@ -562,6 +556,12 @@
       "bytes": 4987866,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.145182983000268,
+      "bytes": 6203180,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -302,3 +302,4 @@
 {"step": 267300, "loss": 0.577639976516366, "lr": 0.0003, "grad_norm": 0.13630385922962587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39124917369, "tokens_per_second": 148490.21348517307, "data_wait_s": 1.5908997738733888e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 267400, "loss": 0.5924610942602158, "lr": 0.0003, "grad_norm": 0.14817470873986413, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39139619055, "tokens_per_second": 149386.27942599004, "data_wait_s": 1.6340996808139607e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 267500, "loss": 0.5844109728932381, "lr": 0.0003, "grad_norm": 0.1551011658965191, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39154049070, "tokens_per_second": 146888.62241168067, "data_wait_s": 1.953000173671171e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 267600, "loss": 0.5919433161616325, "lr": 0.0003, "grad_norm": 0.15521812435714716, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 39168598739, "tokens_per_second": 147657.67878407598, "data_wait_s": 1.7366000975016505e-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:e877f382e01b8e7e29f8da0dd4b607d0e9cc3ef9e0642f3bf5f3c97d57e8f443
-size 648735
+oid sha256:5c00410bdaf816f0f55cb2fe9430c3eb29d67aaf30bcebcf6716bfe8ca4f59e4
+size 650853