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": 77205.69661380946,
-      "required_pages_per_second_upper_estimate": 0.0053412692336404196
+      "required_token_positions_per_second": 77051.28212522679,
+      "required_pages_per_second_upper_estimate": 0.005330586481029294
     },
     {
       "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": 9650.712076726182,
-      "required_pages_per_second_upper_estimate": 0.004040551616740375
+      "required_token_positions_per_second": 9631.410265653349,
+      "required_pages_per_second_upper_estimate": 0.004032470351511829
     },
     {
       "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": 19301.424153452364,
-      "required_pages_per_second_upper_estimate": 0.004019977638379574
+      "required_token_positions_per_second": 19262.820531306697,
+      "required_pages_per_second_upper_estimate": 0.0040119375219326075
     },
     {
       "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": 28952.136230178545,
-      "required_pages_per_second_upper_estimate": 0.001735636359646869
+      "required_token_positions_per_second": 28894.230796960044,
+      "required_pages_per_second_upper_estimate": 0.001732165017341897
     },
     {
       "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": 28952.136230178545,
-      "required_pages_per_second_upper_estimate": 0.00225090124490755
+      "required_token_positions_per_second": 28894.230796960044,
+      "required_pages_per_second_upper_estimate": 0.0022463993521738947
     },
     {
       "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": 28952.136230178545,
-      "required_pages_per_second_upper_estimate": 0.3713669171788272
+      "required_token_positions_per_second": 28894.230796960044,
+      "required_pages_per_second_upper_estimate": 0.3706241684555104
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08611524378136437,
-    "producer_p95_seconds": 0.29751900900009787,
-    "gpu_update_seconds": 0.9549554402545548,
-    "producer_batches_per_second": 11.612345922620536,
-    "gpu_batches_per_second": 1.0471692791586569,
-    "producer_headroom": 11.089272912924278,
+    "producer_batch_seconds": 0.10667324309372361,
+    "producer_p95_seconds": 0.3100237090038718,
+    "gpu_update_seconds": 0.9568692170517596,
+    "producer_batches_per_second": 9.374422029349903,
+    "gpu_batches_per_second": 1.0450748986168998,
+    "producer_headroom": 8.97009586753681,
     "sustainable": true,
-    "buffer_seconds": 60.16219273603695,
+    "buffer_seconds": 60.28276067426086,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362245.4411740977,
-    "upload_bytes_per_second": 5705183.1239266,
+    "generation_bytes_per_second": 448822.28372761805,
+    "upload_bytes_per_second": 5652003.746325318,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.72943903367232,
+    "estimated_full_shards_per_hour": 96.30681403991133,
     "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": 1.3574339080005302,
-      "bytes": 10377609,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.2949794519990974,
       "bytes": 7882340,
@@ -562,6 +556,12 @@
       "bytes": 7357161,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.33853296199959,
+      "bytes": 6323168,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -352,3 +352,4 @@
 {"step": 720200, "loss": 0.5344586372375488, "lr": 8e-05, "grad_norm": 0.22277006736838195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105416991235, "tokens_per_second": 151680.44329242178, "data_wait_s": 1.6111000149976462e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720300, "loss": 0.5282149910926819, "lr": 8e-05, "grad_norm": 0.21542854874343864, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105431760929, "tokens_per_second": 153713.69739467345, "data_wait_s": 1.4101999113336205e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 720400, "loss": 0.5309709310531616, "lr": 8e-05, "grad_norm": 0.2127561535188853, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105446489981, "tokens_per_second": 153689.68463523037, "data_wait_s": 1.5589997929055244e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 720500, "loss": 0.5453026294708252, "lr": 8e-05, "grad_norm": 0.2268957195603477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105461179941, "tokens_per_second": 153552.91413386018, "data_wait_s": 2.015000063693151e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5eeea5b171ad8703e099ac715c5d99c99b4ff36d178190d7e22959d9c2702102
-size 755418
+oid sha256:649fe8acaed083d6a665808f20a171cd21c81cb180d0ee8ce92b902c2672d7ac
+size 757536