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": 75951.60454279944,
-      "required_pages_per_second_upper_estimate": 0.00658329258418804
+      "required_token_positions_per_second": 76108.93433280042,
+      "required_pages_per_second_upper_estimate": 0.006596929531636618
     },
     {
       "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": 9493.95056784993,
-      "required_pages_per_second_upper_estimate": 0.0037872788287258378
+      "required_token_positions_per_second": 9513.616791600052,
+      "required_pages_per_second_upper_estimate": 0.0037951239794160093
     },
     {
       "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": 18987.90113569986,
-      "required_pages_per_second_upper_estimate": 0.00439251675209769
+      "required_token_positions_per_second": 19027.233583200104,
+      "required_pages_per_second_upper_estimate": 0.004401615621599438
     },
     {
       "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": 28481.85170354979,
-      "required_pages_per_second_upper_estimate": 0.0017343953360190634
+      "required_token_positions_per_second": 28540.85037480016,
+      "required_pages_per_second_upper_estimate": 0.0017379880455561055
     },
     {
       "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": 28481.85170354979,
-      "required_pages_per_second_upper_estimate": 0.002192611786950943
+      "required_token_positions_per_second": 28540.85037480016,
+      "required_pages_per_second_upper_estimate": 0.002197153667982571
     },
     {
       "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": 28481.85170354979,
-      "required_pages_per_second_upper_estimate": 0.3536492755323614
+      "required_token_positions_per_second": 28540.85037480016,
+      "required_pages_per_second_upper_estimate": 0.35438184157344027
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06975058502344922,
-    "producer_p95_seconds": 0.19474890299989056,
-    "gpu_update_seconds": 0.970723402669572,
-    "producer_batches_per_second": 14.33679731379763,
-    "gpu_batches_per_second": 1.0301595668239942,
-    "producer_headroom": 13.917064671833616,
+    "producer_batch_seconds": 0.052787744875004705,
+    "producer_p95_seconds": 0.18188725000072736,
+    "gpu_update_seconds": 0.9687167564009063,
+    "producer_batches_per_second": 18.943790881157827,
+    "gpu_batches_per_second": 1.0322934886718809,
+    "producer_headroom": 18.351167656332276,
     "sustainable": true,
-    "buffer_seconds": 60.18485096551346,
+    "buffer_seconds": 60.06043889685619,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323080.20770799613,
-    "upload_bytes_per_second": 5896528.055311814,
+    "generation_bytes_per_second": 349185.84650078014,
+    "upload_bytes_per_second": 5802356.205959822,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.3254916518203,
+    "estimated_full_shards_per_hour": 74.92715402858308,
     "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": [
@@ -424,6 +424,12 @@
       "bytes": 6663126,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9133121940003548,
+      "bytes": 1030486,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 47400, "loss": 0.6294107064604759, "lr": 0.0003, "grad_norm": 0.12717438974419965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6936350677, "tokens_per_second": 150994.34081723812, "data_wait_s": 9.824000699154567e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 47500, "loss": 0.6378660574555397, "lr": 0.0003, "grad_norm": 0.14740369490612742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6951235944, "tokens_per_second": 152434.0333960332, "data_wait_s": 9.100000170292333e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 47600, "loss": 0.6258286442607641, "lr": 0.0003, "grad_norm": 0.1281667609568309, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6965891928, "tokens_per_second": 150372.8449728171, "data_wait_s": 8.529000297130551e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 47700, "loss": 0.644190676510334, "lr": 0.0003, "grad_norm": 0.13149846331555648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6980288879, "tokens_per_second": 145472.13143954135, "data_wait_s": 8.947001333581284e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:647e34c2ad45d78b2546a7035b8979260a0fc457c1bbc701520e989ec92f0d39
-size 161916
+oid sha256:a2dd574b0615284badb3747371ed5d11e21a941db625d863528591724453fcff
+size 164034