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": 75162.28516224607,
-      "required_pages_per_second_upper_estimate": 0.0065148763807933545
+      "required_token_positions_per_second": 75674.49372353523,
+      "required_pages_per_second_upper_estimate": 0.006559273320705168
     },
     {
       "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": 9395.28564528076,
-      "required_pages_per_second_upper_estimate": 0.0037479199159409443
+      "required_token_positions_per_second": 9459.311715441903,
+      "required_pages_per_second_upper_estimate": 0.00377346087260328
     },
     {
       "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": 18790.57129056152,
-      "required_pages_per_second_upper_estimate": 0.004346867965311595
+      "required_token_positions_per_second": 18918.623430883807,
+      "required_pages_per_second_upper_estimate": 0.004376490574334458
     },
     {
       "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": 28185.85693584228,
-      "required_pages_per_second_upper_estimate": 0.0017163708076302028
+      "required_token_positions_per_second": 28377.935146325708,
+      "required_pages_per_second_upper_estimate": 0.0017280673628921567
     },
     {
       "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": 28185.85693584228,
-      "required_pages_per_second_upper_estimate": 0.0021698252903668597
+      "required_token_positions_per_second": 28377.935146325708,
+      "required_pages_per_second_upper_estimate": 0.0021846120027163916
     },
     {
       "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": 28185.85693584228,
-      "required_pages_per_second_upper_estimate": 0.3499740111481962
+      "required_token_positions_per_second": 28377.935146325708,
+      "required_pages_per_second_upper_estimate": 0.3523589796779829
     }
   ],
   "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": 0.486098074601486,
-    "producer_p95_seconds": 0.2824837429998297,
-    "gpu_update_seconds": 0.9809174886161323,
-    "producer_batches_per_second": 2.057198027002724,
-    "gpu_batches_per_second": 1.0194537375521657,
-    "producer_headroom": 2.017941522233574,
+    "producer_batch_seconds": 0.08146859196861556,
+    "producer_p95_seconds": 0.2112628089998907,
+    "gpu_update_seconds": 0.9742780740542986,
+    "producer_batches_per_second": 12.274668996184857,
+    "gpu_batches_per_second": 1.026401010790137,
+    "producer_headroom": 11.958940869256992,
     "sustainable": true,
-    "buffer_seconds": 60.8168842942002,
+    "buffer_seconds": 60.405240591366514,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 204003.87435768588,
-    "upload_bytes_per_second": 6588590.969643014,
+    "generation_bytes_per_second": 216152.68118084158,
+    "upload_bytes_per_second": 6501573.188858819,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 43.77448247001583,
+    "estimated_full_shards_per_hour": 46.38133360451637,
     "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": 1.1758913479998228,
-      "bytes": 13372381,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 1.208357038000031,
       "bytes": 5841536,
@@ -562,6 +556,12 @@
       "bytes": 11119971,
       "pages": 21,
       "shards": 3
+    },
+    {
+      "seconds": 1.229843424000137,
+      "bytes": 7832261,
+      "pages": 45,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 12600, "loss": 0.7019401229918003, "lr": 0.0003, "grad_norm": 0.1562949169520766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1842937690, "tokens_per_second": 148382.08404847415, "data_wait_s": 1.465599962102715e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 12700, "loss": 0.7312840530648828, "lr": 0.0003, "grad_norm": 0.14751725793743778, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1857554894, "tokens_per_second": 148643.9782093223, "data_wait_s": 1.0300000212737359e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 12800, "loss": 0.6953440681099892, "lr": 0.0003, "grad_norm": 0.15038179881078473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1872237247, "tokens_per_second": 133584.30676730064, "data_wait_s": 1.070300095307175e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 12900, "loss": 0.8119467496871948, "lr": 0.0003, "grad_norm": 0.2041011837749955, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1886921849, "tokens_per_second": 150302.32083414425, "data_wait_s": 1.843299833126366e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0e1fefc40e347e7fd7e315bc92ac0610bbff951d48ad492c47a442c8429c71d
-size 268296
+oid sha256:a059ceb301cf6aacd1590cc15c28db911057b71bebe0a5445d378c09f931370a
+size 270379