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": 75974.34292901409,
-      "required_pages_per_second_upper_estimate": 0.006585263489875213
+      "required_token_positions_per_second": 76079.09180030337,
+      "required_pages_per_second_upper_estimate": 0.006594342856554981
     },
     {
       "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": 9496.792866126762,
-      "required_pages_per_second_upper_estimate": 0.0037884126640046123
+      "required_token_positions_per_second": 9509.886475037922,
+      "required_pages_per_second_upper_estimate": 0.0037936359003661727
     },
     {
       "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": 18993.585732253523,
-      "required_pages_per_second_upper_estimate": 0.004393831783464897
+      "required_token_positions_per_second": 19019.772950075843,
+      "required_pages_per_second_upper_estimate": 0.004399889735428798
     },
     {
       "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": 28490.378598380288,
-      "required_pages_per_second_upper_estimate": 0.001734914579177081
+      "required_token_positions_per_second": 28529.659425113765,
+      "required_pages_per_second_upper_estimate": 0.0017373065754345794
     },
     {
       "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": 28490.378598380288,
-      "required_pages_per_second_upper_estimate": 0.0021932682109189502
+      "required_token_positions_per_second": 28529.659425113765,
+      "required_pages_per_second_upper_estimate": 0.002196292157697181
     },
     {
       "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": 28490.378598380288,
-      "required_pages_per_second_upper_estimate": 0.353755151028475
+      "required_token_positions_per_second": 28529.659425113765,
+      "required_pages_per_second_upper_estimate": 0.3542428874320345
     }
   ],
   "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": 0.05513317221105751,
-    "producer_p95_seconds": 0.18817629399927682,
-    "gpu_update_seconds": 0.9704328745414363,
-    "producer_batches_per_second": 18.13790064848545,
-    "gpu_batches_per_second": 1.0304679759252129,
-    "producer_headroom": 17.601615064456716,
+    "producer_batch_seconds": 0.10278326035158614,
+    "producer_p95_seconds": 0.19477145499877224,
+    "gpu_update_seconds": 0.9690967420263816,
+    "producer_batches_per_second": 9.729210735088033,
+    "gpu_batches_per_second": 1.0318887234199132,
+    "producer_headroom": 9.42854642586191,
     "sustainable": true,
-    "buffer_seconds": 60.16683822156905,
+    "buffer_seconds": 60.083998005635664,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291628.81761863903,
-    "upload_bytes_per_second": 5433723.454935909,
+    "generation_bytes_per_second": 315719.7560790267,
+    "upload_bytes_per_second": 5561178.492265481,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.57675548953417,
+    "estimated_full_shards_per_hour": 67.74611007478809,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 592565,
       "pages": 19,
       "shards": 1
+    },
+    {
+      "seconds": 1.3723258429999987,
+      "bytes": 11921886,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -35,3 +35,4 @@
 {"step": 283700, "loss": 0.6043622884899378, "lr": 0.0003, "grad_norm": 0.13929980521587076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41525804482, "tokens_per_second": 150126.03680142402, "data_wait_s": 1.4197999917087145e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 283800, "loss": 0.5926853604614735, "lr": 0.0003, "grad_norm": 0.15031356883188096, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41540394660, "tokens_per_second": 134554.33722788843, "data_wait_s": 1.3687998944078572e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 283900, "loss": 0.534436808899045, "lr": 0.0003, "grad_norm": 0.16632695571089925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41555084167, "tokens_per_second": 150336.53961063953, "data_wait_s": 1.3389999367063865e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 284000, "loss": 0.5654576811939478, "lr": 0.0003, "grad_norm": 0.14999594140498718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41569809586, "tokens_per_second": 150860.31564225064, "data_wait_s": 2.1712998204748146e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9632bbb783d6e2b46ed897b5d08988033448d2fe5e49872b93386a8f3f07df1f
-size 78531
+oid sha256:d3cbaa3cbe6f6ad8de0c4ba36c1a8eb4212d80a93c264a6f136885404a3edfbb
+size 80649