Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76407.27063638637,
-      "required_pages_per_second_upper_estimate": 0.005286032271918783
+      "required_token_positions_per_second": 76537.65259061333,
+      "required_pages_per_second_upper_estimate": 0.005295052398040018
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9550.908829548296,
-      "required_pages_per_second_upper_estimate": 0.003998766081275789
+      "required_token_positions_per_second": 9567.206573826666,
+      "required_pages_per_second_upper_estimate": 0.004005589606469542
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19101.817659096592,
-      "required_pages_per_second_upper_estimate": 0.00397840486958251
+      "required_token_positions_per_second": 19134.413147653333,
+      "required_pages_per_second_upper_estimate": 0.00398519365024804
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28652.72648864489,
-      "required_pages_per_second_upper_estimate": 0.0017176872028141306
+      "required_token_positions_per_second": 28701.619721479998,
+      "required_pages_per_second_upper_estimate": 0.0017206182774669515
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28652.72648864489,
-      "required_pages_per_second_upper_estimate": 0.00222762345446758
+      "required_token_positions_per_second": 28701.619721479998,
+      "required_pages_per_second_upper_estimate": 0.0022314246882619067
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28652.72648864489,
-      "required_pages_per_second_upper_estimate": 0.3675264104955669
+      "required_token_positions_per_second": 28701.619721479998,
+      "required_pages_per_second_upper_estimate": 0.3681535603889124
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04663595294531575,
-    "producer_p95_seconds": 0.13081054999997832,
-    "gpu_update_seconds": 0.964934349649306,
-    "producer_batches_per_second": 21.442683956143817,
-    "gpu_batches_per_second": 1.036339933761751,
-    "producer_headroom": 20.69078229795724,
+    "producer_batch_seconds": 0.05992200722656449,
+    "producer_p95_seconds": 0.13660705500001313,
+    "gpu_update_seconds": 0.9632905831899278,
+    "producer_batches_per_second": 16.688359524056168,
+    "gpu_batches_per_second": 1.0381083521947336,
+    "producer_headroom": 16.075739578411252,
     "sustainable": true,
-    "buffer_seconds": 60.79086402790628,
+    "buffer_seconds": 60.68730674096545,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 67,
+    "upload_pages_per_commit": 68,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432625.8109084137,
-    "upload_bytes_per_second": 3017589.0709228506,
+    "generation_bytes_per_second": 465348.27414554404,
+    "upload_bytes_per_second": 3122923.41202882,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.83142800750073,
+    "estimated_full_shards_per_hour": 99.85290687823048,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 12173751,
       "pages": 40,
       "shards": 4
+    },
+    {
+      "seconds": 1.4142575109999598,
+      "bytes": 6051595,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -9,3 +9,4 @@
 {"step": 843400, "loss": 0.4981638491153717, "lr": 3e-05, "grad_norm": 0.30460754308575466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123450026610, "tokens_per_second": 151846.02516218185, "data_wait_s": 1.2015000038445578e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843500, "loss": 0.5259416103363037, "lr": 3e-05, "grad_norm": 0.2947109431887774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123464715616, "tokens_per_second": 151901.25280813174, "data_wait_s": 1.4360000022861641e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 843600, "loss": 0.5174587368965149, "lr": 3e-05, "grad_norm": 0.29567694041875386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123479457093, "tokens_per_second": 152771.79248379884, "data_wait_s": 1.0048000149254221e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 843700, "loss": 0.5038239359855652, "lr": 3e-05, "grad_norm": 0.2985374037328828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123494154041, "tokens_per_second": 130223.20923862039, "data_wait_s": 1.32210000174382e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee20da60e6e3ad0bb3a9f74f157ad8d60144758c83fc12ef5af3007b7dd8bef6
-size 23463
+oid sha256:b4c28b15671aa669191e6dfa87bd78dc18cb1a9665e6ed172bcc0be6ec236292
+size 25581