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": 76204.77791046996,
-      "required_pages_per_second_upper_estimate": 0.005272023355292029
+      "required_token_positions_per_second": 76712.5676302114,
+      "required_pages_per_second_upper_estimate": 0.005307153426338221
     },
     {
       "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": 9525.597238808745,
-      "required_pages_per_second_upper_estimate": 0.0039881686467992585
+      "required_token_positions_per_second": 9589.070953776425,
+      "required_pages_per_second_upper_estimate": 0.004014743765774332
     },
     {
       "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": 19051.19447761749,
-      "required_pages_per_second_upper_estimate": 0.003967861395903485
+      "required_token_positions_per_second": 19178.14190755285,
+      "required_pages_per_second_upper_estimate": 0.003994301197730161
     },
     {
       "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": 28576.791716426233,
-      "required_pages_per_second_upper_estimate": 0.0017131350291653058
+      "required_token_positions_per_second": 28767.212861329273,
+      "required_pages_per_second_upper_estimate": 0.0017245504860459902
     },
     {
       "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": 28576.791716426233,
-      "required_pages_per_second_upper_estimate": 0.0022217198599293443
+      "required_token_positions_per_second": 28767.212861329273,
+      "required_pages_per_second_upper_estimate": 0.0022365242663714567
     },
     {
       "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": 28576.791716426233,
-      "required_pages_per_second_upper_estimate": 0.36655240076995205
+      "required_token_positions_per_second": 28767.212861329273,
+      "required_pages_per_second_upper_estimate": 0.3689949187584725
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.48517990075777107,
-    "producer_p95_seconds": 0.31256719500015606,
-    "gpu_update_seconds": 0.9674983908045789,
-    "producer_batches_per_second": 2.061091150804402,
-    "gpu_batches_per_second": 1.0335934503915738,
-    "producer_headroom": 1.9941023717048167,
+    "producer_batch_seconds": 0.09997281796091784,
+    "producer_p95_seconds": 0.29959333799706656,
+    "gpu_update_seconds": 0.9610941502493002,
+    "producer_batches_per_second": 10.002718942972358,
+    "gpu_batches_per_second": 1.0404807892552543,
+    "producer_headroom": 9.613554662678597,
     "sustainable": true,
-    "buffer_seconds": 60.95239862068847,
+    "buffer_seconds": 60.54893146570591,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318406.7806369972,
-    "upload_bytes_per_second": 5852692.165549001,
+    "generation_bytes_per_second": 306647.07262399414,
+    "upload_bytes_per_second": 5944504.25387808,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.3226829941982,
+    "estimated_full_shards_per_hour": 65.7993234065997,
     "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": 1.1072141639997426,
-      "bytes": 6205762,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 0.9200315849993785,
       "bytes": 2298271,
@@ -562,6 +556,12 @@
       "bytes": 9840833,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.1696946029987885,
+      "bytes": 13219787,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -299,3 +299,4 @@
 {"step": 427000, "loss": 0.5588080286979675, "lr": 0.0003, "grad_norm": 0.15673533564415312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62500857386, "tokens_per_second": 150635.67784604905, "data_wait_s": 1.0767998901428655e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427100, "loss": 0.5821868181228638, "lr": 0.0003, "grad_norm": 0.1780204152947438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62515472162, "tokens_per_second": 150749.37400461477, "data_wait_s": 1.3649001630255952e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427200, "loss": 0.5943984389305115, "lr": 0.0003, "grad_norm": 0.14836201436518875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62530168164, "tokens_per_second": 152614.19916724236, "data_wait_s": 1.2024000170640647e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 427300, "loss": 0.5756198167800903, "lr": 0.0003, "grad_norm": 0.168894812228803, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62544784988, "tokens_per_second": 152351.00901860333, "data_wait_s": 1.0269999620504677e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b6e95e340161c5360298001615a88bfb2478c6eb043f4a5f3991f0545880bc2
-size 642381
+oid sha256:39d834e5148cf0fda0fb25c78840495d5432cc41f3441de7871d46c0cf1ac022
+size 644499