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": 74490.46556253861,
-      "required_pages_per_second_upper_estimate": 0.00645664476060192
+      "required_token_positions_per_second": 74572.72551024861,
+      "required_pages_per_second_upper_estimate": 0.006463774844383491
     },
     {
       "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": 9311.308195317326,
-      "required_pages_per_second_upper_estimate": 0.0037144200555757644
+      "required_token_positions_per_second": 9321.590688781076,
+      "required_pages_per_second_upper_estimate": 0.0037185218959554316
     },
     {
       "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": 18622.616390634652,
-      "required_pages_per_second_upper_estimate": 0.004308014555118796
+      "required_token_positions_per_second": 18643.181377562152,
+      "required_pages_per_second_upper_estimate": 0.0043127719029130655
     },
     {
       "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": 27933.92458595198,
-      "required_pages_per_second_upper_estimate": 0.0017010294493087703
+      "required_token_positions_per_second": 27964.772066343226,
+      "required_pages_per_second_upper_estimate": 0.0017029078990203755
     },
     {
       "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": 27933.92458595198,
-      "required_pages_per_second_upper_estimate": 0.002150430840678925
+      "required_token_positions_per_second": 27964.772066343226,
+      "required_pages_per_second_upper_estimate": 0.002152805565110734
     },
     {
       "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": 27933.92458595198,
-      "required_pages_per_second_upper_estimate": 0.34684585452589467
+      "required_token_positions_per_second": 27964.772066343226,
+      "required_pages_per_second_upper_estimate": 0.34722887699247834
     }
   ],
   "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": 0.06536761660152735,
-    "producer_p95_seconds": 0.17454405199987377,
-    "gpu_update_seconds": 0.9897642529580046,
-    "producer_batches_per_second": 15.298094867002304,
-    "gpu_batches_per_second": 1.0103416010543975,
-    "producer_headroom": 15.14150743771922,
+    "producer_batch_seconds": 0.0502049259921904,
+    "producer_p95_seconds": 0.17078043600031378,
+    "gpu_update_seconds": 0.988672460279965,
+    "producer_batches_per_second": 19.91836418911472,
+    "gpu_batches_per_second": 1.0114573230014188,
+    "producer_headroom": 19.6927381276044,
     "sustainable": true,
-    "buffer_seconds": 60.375619430438284,
+    "buffer_seconds": 60.30902007707787,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422320.1799184866,
-    "upload_bytes_per_second": 5051501.2185256425,
+    "generation_bytes_per_second": 419946.2114562384,
+    "upload_bytes_per_second": 5143518.243111111,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.62007949987363,
+    "estimated_full_shards_per_hour": 90.1106811310326,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 7166128,
       "pages": 34,
       "shards": 4
+    },
+    {
+      "seconds": 1.0863462409997737,
+      "bytes": 6279325,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -36,3 +36,4 @@
 {"step": 198600, "loss": 0.6101530659943819, "lr": 0.0003, "grad_norm": 0.14258927592460816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29067243499, "tokens_per_second": 148175.9281633369, "data_wait_s": 1.7857999409898184e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198700, "loss": 0.5925858672708273, "lr": 0.0003, "grad_norm": 0.1401057552846532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29081927637, "tokens_per_second": 148369.93185458748, "data_wait_s": 1.7082000340451486e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198800, "loss": 0.5626566894352436, "lr": 0.0003, "grad_norm": 0.1311758286935325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29096534059, "tokens_per_second": 147478.5739769078, "data_wait_s": 2.369800040469272e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 198900, "loss": 0.5834132824093103, "lr": 0.0003, "grad_norm": 0.15648794412353406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29111204742, "tokens_per_second": 147019.28082561592, "data_wait_s": 2.321600004506763e-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 sha256:26a2d8d0047a53297bac3e30e5098b565e6b251373a1d1f5741e5696711cd269
-size 80649
+oid sha256:fea717f6c89919c0d2333dbab2ceee4b30e6b5de823ab84b114990f2219356ea
+size 82767