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": 75792.17806812613,
-      "required_pages_per_second_upper_estimate": 0.006569473901426063
+      "required_token_positions_per_second": 75838.67028275809,
+      "required_pages_per_second_upper_estimate": 0.006573503728756921
     },
     {
       "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": 9474.022258515766,
-      "required_pages_per_second_upper_estimate": 0.003779329128177663
+      "required_token_positions_per_second": 9479.833785344761,
+      "required_pages_per_second_upper_estimate": 0.0037816474331198186
     },
     {
       "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": 18948.044517031532,
-      "required_pages_per_second_upper_estimate": 0.004383296624821303
+      "required_token_positions_per_second": 18959.667570689522,
+      "required_pages_per_second_upper_estimate": 0.004385985413726322
     },
     {
       "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": 28422.0667755473,
-      "required_pages_per_second_upper_estimate": 0.0017307547475709875
+      "required_token_positions_per_second": 28439.50135603428,
+      "required_pages_per_second_upper_estimate": 0.00173181642205047
     },
     {
       "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": 28422.0667755473,
-      "required_pages_per_second_upper_estimate": 0.002188009377698038
+      "required_token_positions_per_second": 28439.50135603428,
+      "required_pages_per_second_upper_estimate": 0.0021893515399659335
     },
     {
       "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": 28422.0667755473,
-      "required_pages_per_second_upper_estimate": 0.35290694681385326
+      "required_token_positions_per_second": 28439.50135603428,
+      "required_pages_per_second_upper_estimate": 0.3531234259537142
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.05000970483592937,
-    "producer_p95_seconds": 0.1727249159994244,
-    "gpu_update_seconds": 0.9727652889685959,
-    "producer_batches_per_second": 19.99611881895276,
-    "gpu_batches_per_second": 1.0279972068702004,
-    "producer_headroom": 19.451530301168958,
+    "producer_batch_seconds": 0.09408916248436583,
+    "producer_p95_seconds": 0.19224768100048095,
+    "gpu_update_seconds": 0.9721689439584235,
+    "producer_batches_per_second": 10.628216614917402,
+    "gpu_batches_per_second": 1.0286277978889715,
+    "producer_headroom": 10.332422122685621,
     "sustainable": true,
-    "buffer_seconds": 60.31144791605295,
+    "buffer_seconds": 60.27447452542226,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 439633.66905388614,
-    "upload_bytes_per_second": 4798111.443481281,
+    "generation_bytes_per_second": 468715.5925343753,
+    "upload_bytes_per_second": 4826734.250837589,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.33515123093069,
+    "estimated_full_shards_per_hour": 100.57545501731343,
     "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": [
@@ -388,6 +388,12 @@
       "bytes": 6320844,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.26741677300015,
+      "bytes": 7410356,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -40,3 +40,4 @@
 {"step": 241100, "loss": 0.6115792747586966, "lr": 0.0003, "grad_norm": 0.14419259800452822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35289707958, "tokens_per_second": 149590.5110744044, "data_wait_s": 1.5069000255607534e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 241200, "loss": 0.6091801077127457, "lr": 0.0003, "grad_norm": 0.1659018137984772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35304302650, "tokens_per_second": 147903.99893654566, "data_wait_s": 1.2401000276440755e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 241300, "loss": 0.6611749287694693, "lr": 0.0003, "grad_norm": 0.15601242358858483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35319001314, "tokens_per_second": 150003.28089799767, "data_wait_s": 1.2620000234164763e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 241400, "loss": 0.6207055300474167, "lr": 0.0003, "grad_norm": 0.13743057772118955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35333684821, "tokens_per_second": 149884.36847653115, "data_wait_s": 1.9915999473596457e-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:d7f51dfca8c5ad3efa7dd5a463a00d897c0e20aac3ab331a487fedd8355242b7
-size 89904
+oid sha256:9c3b2e540a337da3c473a107048183cdbe6592674c890eb93fb4d5c4bdb5a0d9
+size 92022