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": 76140.58584390576,
-      "required_pages_per_second_upper_estimate": 0.006599673004399208
+      "required_token_positions_per_second": 75396.74936134786,
+      "required_pages_per_second_upper_estimate": 0.006535199143327418
     },
     {
       "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": 9517.57323048822,
-      "required_pages_per_second_upper_estimate": 0.0037967022620425325
+      "required_token_positions_per_second": 9424.593670168482,
+      "required_pages_per_second_upper_estimate": 0.0037596113252626783
     },
     {
       "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": 19035.14646097644,
-      "required_pages_per_second_upper_estimate": 0.004403446126611101
+      "required_token_positions_per_second": 18849.187340336965,
+      "required_pages_per_second_upper_estimate": 0.004360427756819902
     },
     {
       "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": 28552.71969146466,
-      "required_pages_per_second_upper_estimate": 0.0017387108246674817
+      "required_token_positions_per_second": 28273.781010505445,
+      "required_pages_per_second_upper_estimate": 0.001721724922475221
     },
     {
       "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": 28552.71969146466,
-      "required_pages_per_second_upper_estimate": 0.0021980674008357758
+      "required_token_positions_per_second": 28273.781010505445,
+      "required_pages_per_second_upper_estimate": 0.0021765939290238467
     },
     {
       "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": 28552.71969146466,
-      "required_pages_per_second_upper_estimate": 0.35452921876236587
+      "required_token_positions_per_second": 28273.781010505445,
+      "required_pages_per_second_upper_estimate": 0.35106573389256424
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09273642833593954,
-    "producer_p95_seconds": 0.3786180819988658,
-    "gpu_update_seconds": 0.9683140625046969,
-    "producer_batches_per_second": 10.78324902030387,
-    "gpu_batches_per_second": 1.032722789766517,
-    "producer_headroom": 10.441571665850232,
+    "producer_batch_seconds": 0.11331817983585779,
+    "producer_p95_seconds": 0.3868185720020847,
+    "gpu_update_seconds": 0.9778670914133157,
+    "producer_batches_per_second": 8.824709340094477,
+    "gpu_batches_per_second": 1.0226338617804343,
+    "producer_headroom": 8.629392854966108,
     "sustainable": true,
-    "buffer_seconds": 60.03547187529121,
+    "buffer_seconds": 60.627759667625575,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 490604.41807994706,
-    "upload_bytes_per_second": 5193658.35176837,
+    "generation_bytes_per_second": 507609.88189362624,
+    "upload_bytes_per_second": 5244076.132598249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.2722874336129,
+    "estimated_full_shards_per_hour": 108.9212641010913,
     "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": 0.8281688100032625,
-      "bytes": 661491,
-      "pages": 19,
-      "shards": 1
-    },
     {
       "seconds": 1.42470605500057,
       "bytes": 11031230,
@@ -562,6 +556,12 @@
       "bytes": 7600114,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.056514903997595,
+      "bytes": 5902748,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -323,3 +323,4 @@
 {"step": 147300, "loss": 0.6135005541145802, "lr": 0.0003, "grad_norm": 0.14085108673209615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21556893328, "tokens_per_second": 150064.4933955498, "data_wait_s": 1.6808997315820307e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147400, "loss": 0.577962338924408, "lr": 0.0003, "grad_norm": 0.14103880552126036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21571622651, "tokens_per_second": 150578.31751383236, "data_wait_s": 1.5854002413107082e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147500, "loss": 0.6346491239964962, "lr": 0.0003, "grad_norm": 0.14649534405941178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21586278425, "tokens_per_second": 149279.80276100218, "data_wait_s": 1.326200072071515e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 147600, "loss": 0.6685522776097059, "lr": 0.0003, "grad_norm": 0.15143513756810176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21600793368, "tokens_per_second": 148912.81795938374, "data_wait_s": 1.0970994480885565e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86c760f66ea1c24b9ed207c40a3778d7eb0bdcffd99c83dac1ba47804d47f313
-size 693213
+oid sha256:aa2f330f8db1a3b8ac22170f76a52a5a4fa9db5ac461a4e1935078cf1cba9b3b
+size 695331