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": 74522.1844950873,
-      "required_pages_per_second_upper_estimate": 0.0064593940772843415
+      "required_token_positions_per_second": 74457.28901019893,
+      "required_pages_per_second_upper_estimate": 0.00645376910113032
     },
     {
       "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": 9315.273061885913,
-      "required_pages_per_second_upper_estimate": 0.0037160017001300117
+      "required_token_positions_per_second": 9307.161126274867,
+      "required_pages_per_second_upper_estimate": 0.0037127657277305198
     },
     {
       "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": 18630.546123771826,
-      "required_pages_per_second_upper_estimate": 0.004309848959321546
+      "required_token_positions_per_second": 18614.322252549733,
+      "required_pages_per_second_upper_estimate": 0.004306095852244695
     },
     {
       "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": 27945.81918565774,
-      "required_pages_per_second_upper_estimate": 0.0017017537680247094
+      "required_token_positions_per_second": 27921.483378824603,
+      "required_pages_per_second_upper_estimate": 0.001700271845068682
     },
     {
       "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": 27945.81918565774,
-      "required_pages_per_second_upper_estimate": 0.0021513465198906337
+      "required_token_positions_per_second": 27921.483378824603,
+      "required_pages_per_second_upper_estimate": 0.002149473082114794
     },
     {
       "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": 27945.81918565774,
-      "required_pages_per_second_upper_estimate": 0.34699354564557583
+      "required_token_positions_per_second": 27921.483378824603,
+      "required_pages_per_second_upper_estimate": 0.3466913763714144
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12180777489055572,
-    "producer_p95_seconds": 0.21463039799891703,
-    "gpu_update_seconds": 0.989342978866385,
-    "producer_batches_per_second": 8.209656574865603,
-    "gpu_batches_per_second": 1.0107718166108846,
-    "producer_headroom": 8.122166091247538,
+    "producer_batch_seconds": 0.05750933145296244,
+    "producer_p95_seconds": 0.19648339900049905,
+    "gpu_update_seconds": 0.9902052704323006,
+    "producer_batches_per_second": 17.38848243815722,
+    "gpu_batches_per_second": 1.0098916152642,
+    "producer_headroom": 17.21816695508278,
     "sustainable": true,
-    "buffer_seconds": 60.349921710849486,
+    "buffer_seconds": 60.40252149637033,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374661.8467613629,
+    "generation_bytes_per_second": 364979.9277124252,
     "upload_bytes_per_second": 5332426.758426513,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.39371063357035,
+    "estimated_full_shards_per_hour": 78.31619618920867,
     "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": [

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -93,3 +93,4 @@
 {"step": 204300, "loss": 0.6236441843211651, "lr": 0.0003, "grad_norm": 0.1589290398850428, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29902201561, "tokens_per_second": 149592.67256207936, "data_wait_s": 1.896799949463457e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 204400, "loss": 0.5921995509415865, "lr": 0.0003, "grad_norm": 0.1338520833183724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29916688221, "tokens_per_second": 146192.87678636564, "data_wait_s": 1.921399962157011e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 204500, "loss": 0.6185578256845474, "lr": 0.0003, "grad_norm": 0.15167859882647206, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29931197138, "tokens_per_second": 146159.02259195168, "data_wait_s": 1.7411999579053372e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 204600, "loss": 0.580900875851512, "lr": 0.0003, "grad_norm": 0.15509897919318952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29945759516, "tokens_per_second": 147130.26887192292, "data_wait_s": 2.7001999114872888e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a4239385e44df746ea6cca5c72118b62a07e15a39610bd37aec826341959f94
-size 202158
+oid sha256:eea1e88ba9a8df1dd06dd04b097eb989971487df3b3ee5e76cd39829ef032d2a
+size 204276