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": 72923.36135073398,
-      "required_pages_per_second_upper_estimate": 0.0063208121393120075
+      "required_token_positions_per_second": 73148.75914216327,
+      "required_pages_per_second_upper_estimate": 0.006340349048607619
     },
     {
       "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": 9115.420168841747,
-      "required_pages_per_second_upper_estimate": 0.0036362773930276636
+      "required_token_positions_per_second": 9143.594892770408,
+      "required_pages_per_second_upper_estimate": 0.0036475167116524687
     },
     {
       "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": 18230.840337683494,
-      "required_pages_per_second_upper_estimate": 0.004217384060291584
+      "required_token_positions_per_second": 18287.189785540817,
+      "required_pages_per_second_upper_estimate": 0.004230419513336965
     },
     {
       "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": 27346.26050652524,
-      "required_pages_per_second_upper_estimate": 0.0016652437901068216
+      "required_token_positions_per_second": 27430.78467831122,
+      "required_pages_per_second_upper_estimate": 0.0016703908687045312
     },
     {
       "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": 27346.26050652524,
-      "required_pages_per_second_upper_estimate": 0.0021051908330863654
+      "required_token_positions_per_second": 27430.78467831122,
+      "required_pages_per_second_upper_estimate": 0.002111697737808333
     },
     {
       "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": 27346.26050652524,
-      "required_pages_per_second_upper_estimate": 0.3395490334445688
+      "required_token_positions_per_second": 27430.78467831122,
+      "required_pages_per_second_upper_estimate": 0.3405985407739452
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.10529878528149084,
-    "producer_p95_seconds": 0.43309841900190804,
-    "gpu_update_seconds": 1.0110340312673742,
-    "producer_batches_per_second": 9.49678571625249,
-    "gpu_batches_per_second": 0.9890863898482798,
-    "producer_headroom": 9.601573546785172,
+    "producer_batch_seconds": 0.5154124498902775,
+    "producer_p95_seconds": 0.45972732399968663,
+    "gpu_update_seconds": 1.0079186696347233,
+    "producer_batches_per_second": 1.940193722935647,
+    "gpu_batches_per_second": 0.9921435430523446,
+    "producer_headroom": 1.9555574760549383,
     "sustainable": true,
-    "buffer_seconds": 60.66204187604245,
+    "buffer_seconds": 60.4751201780834,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 293493.73707738944,
-    "upload_bytes_per_second": 5493555.94485042,
+    "generation_bytes_per_second": 269798.27542300324,
+    "upload_bytes_per_second": 5431092.984296833,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.97692379227888,
+    "estimated_full_shards_per_hour": 57.892429323364,
     "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": 1.0272974420004175,
-      "bytes": 6127988,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1608434069930809,
       "bytes": 13413721,
@@ -562,6 +556,12 @@
       "bytes": 13303337,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 0.9868402909996803,
+      "bytes": 1096520,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -423,3 +423,4 @@
 {"step": 357300, "loss": 0.580042626708746, "lr": 0.0003, "grad_norm": 0.156638364723009, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52298628386, "tokens_per_second": 147576.94735376287, "data_wait_s": 1.0554002074059099e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357400, "loss": 0.6024862378835678, "lr": 0.0003, "grad_norm": 0.17464307222932632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52313238998, "tokens_per_second": 148470.39450399738, "data_wait_s": 8.96799610927701e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 357500, "loss": 0.6064887754619122, "lr": 0.0003, "grad_norm": 0.15418517736130338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52327985290, "tokens_per_second": 147869.70254859925, "data_wait_s": 1.3831006071995944e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 357600, "loss": 0.5445334520190954, "lr": 0.0003, "grad_norm": 0.16050267865001872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52342744877, "tokens_per_second": 150226.91947575184, "data_wait_s": 1.571000029798597e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b45017d561eded48e97476b498770c738e67e2645e32b870e3f3f85a2742223
-size 906579
+oid sha256:f1eb6deda87a1e1088f537c558bde22f4c84157e4386575169ef1e1b7f1e012a
+size 908697