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": 75451.49724323375,
-      "required_pages_per_second_upper_estimate": 0.006539944550972049
+      "required_token_positions_per_second": 75207.38750719558,
+      "required_pages_per_second_upper_estimate": 0.006518785737742731
     },
     {
       "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": 9431.437155404219,
-      "required_pages_per_second_upper_estimate": 0.0037623412938424363
+      "required_token_positions_per_second": 9400.923438399448,
+      "required_pages_per_second_upper_estimate": 0.0037501689159085082
     },
     {
       "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": 18862.874310808438,
-      "required_pages_per_second_upper_estimate": 0.004363593996556025
+      "required_token_positions_per_second": 18801.846876798896,
+      "required_pages_per_second_upper_estimate": 0.004349476373744073
     },
     {
       "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": 28294.311466212657,
-      "required_pages_per_second_upper_estimate": 0.0017229751195128138
+      "required_token_positions_per_second": 28202.77031519834,
+      "required_pages_per_second_upper_estimate": 0.0017174007436953434
     },
     {
       "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": 28294.311466212657,
-      "required_pages_per_second_upper_estimate": 0.0021781744203361268
+      "required_token_positions_per_second": 28202.77031519834,
+      "required_pages_per_second_upper_estimate": 0.00217112732912892
     },
     {
       "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": 28294.311466212657,
-      "required_pages_per_second_upper_estimate": 0.3513206534414326
+      "required_token_positions_per_second": 28202.77031519834,
+      "required_pages_per_second_upper_estimate": 0.3501840187143591
     }
   ],
   "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": 0.11120880872695693,
-    "producer_p95_seconds": 0.3540053260003333,
-    "gpu_update_seconds": 0.9771575474814277,
-    "producer_batches_per_second": 8.992093445180487,
-    "gpu_batches_per_second": 1.0233764274527148,
-    "producer_headroom": 8.786691977616387,
+    "producer_batch_seconds": 0.11631730684376862,
+    "producer_p95_seconds": 0.32116989900168846,
+    "gpu_update_seconds": 0.9803292262072787,
+    "producer_batches_per_second": 8.59717291549011,
+    "gpu_batches_per_second": 1.0200654772568845,
+    "producer_headroom": 8.428059871812593,
     "sustainable": true,
-    "buffer_seconds": 60.583767943848514,
+    "buffer_seconds": 60.78041202485128,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347352.3575189596,
+    "generation_bytes_per_second": 347866.14152528474,
     "upload_bytes_per_second": 4483229.491392168,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.53372997452347,
+    "estimated_full_shards_per_hour": 74.64397606200129,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -274,3 +274,4 @@
 {"step": 264500, "loss": 0.6122810952365398, "lr": 0.0003, "grad_norm": 0.18052651250036417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38715760764, "tokens_per_second": 148749.38149347337, "data_wait_s": 1.2296000932110474e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 264600, "loss": 0.555720204487443, "lr": 0.0003, "grad_norm": 0.1334002291754361, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38730258365, "tokens_per_second": 136356.04386173695, "data_wait_s": 1.2864999007433653e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 264700, "loss": 0.5824983902275562, "lr": 0.0003, "grad_norm": 0.16188175198946875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38744856079, "tokens_per_second": 148505.21341425242, "data_wait_s": 1.5558001905446872e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 264800, "loss": 0.5940949842333794, "lr": 0.0003, "grad_norm": 0.15285461701734704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38759615411, "tokens_per_second": 150145.6373441198, "data_wait_s": 1.5038000128697604e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:adfbafc8b7cc1030879d490935f7e1a99b1e93b1828c3e7e1f4c7f88338cc8ea
-size 588648
+oid sha256:529e96067f2ee814a73e97789ead2679a7fe5c3d260deb88c88b4a83d6e4f258
+size 590766