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": 74428.44008614862,
-      "required_pages_per_second_upper_estimate": 0.006451268549510569
+      "required_token_positions_per_second": 75238.62857022154,
+      "required_pages_per_second_upper_estimate": 0.00652149363390607
     },
     {
       "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": 9303.555010768578,
-      "required_pages_per_second_upper_estimate": 0.0037113271943388293
+      "required_token_positions_per_second": 9404.828571277692,
+      "required_pages_per_second_upper_estimate": 0.0037517267318005793
     },
     {
       "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": 18607.110021537155,
-      "required_pages_per_second_upper_estimate": 0.004304427429530866
+      "required_token_positions_per_second": 18809.657142555385,
+      "required_pages_per_second_upper_estimate": 0.004351283141270854
     },
     {
       "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": 27910.66503230573,
-      "required_pages_per_second_upper_estimate": 0.0016996130645251604
+      "required_token_positions_per_second": 28214.485713833074,
+      "required_pages_per_second_upper_estimate": 0.0017181141500062522
     },
     {
       "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": 27910.66503230573,
-      "required_pages_per_second_upper_estimate": 0.002148640255852671
+      "required_token_positions_per_second": 28214.485713833074,
+      "required_pages_per_second_upper_estimate": 0.0021720292129461206
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27910.66503230573,
-      "required_pages_per_second_upper_estimate": 0.3465570487143267
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28214.485713833074,
+      "required_pages_per_second_upper_estimate": 0.3503294847564855
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2550642762423081,
-    "producer_p95_seconds": 0.3604272789998504,
-    "gpu_update_seconds": 0.9905890801239704,
-    "producer_batches_per_second": 3.9205803914696844,
-    "gpu_batches_per_second": 1.0095003266892988,
-    "producer_headroom": 3.8836841235380306,
+    "producer_batch_seconds": 0.10536246880482736,
+    "producer_p95_seconds": 0.3503805110012763,
+    "gpu_update_seconds": 0.9799221676560513,
+    "producer_batches_per_second": 9.491045638389439,
+    "gpu_batches_per_second": 1.0204892112931523,
+    "producer_headroom": 9.30048601529309,
     "sustainable": true,
-    "buffer_seconds": 60.425933887562195,
+    "buffer_seconds": 60.75517439467518,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 455977.7528602481,
-    "upload_bytes_per_second": 5836343.046279267,
+    "generation_bytes_per_second": 481979.696200671,
+    "upload_bytes_per_second": 5848164.134059681,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.8422111449774,
+    "estimated_full_shards_per_hour": 103.42162289157007,
     "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": 1.1489640459985822,
-      "bytes": 6010015,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.3807983080005215,
       "bytes": 15814135,
@@ -562,6 +556,12 @@
       "bytes": 8139088,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.03730981300032,
+      "bytes": 6215262,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -261,3 +261,4 @@
 {"step": 26100, "loss": 0.6592318769544363, "lr": 0.0003, "grad_norm": 0.1358972821967768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3819172841, "tokens_per_second": 149150.48024003956, "data_wait_s": 1.3214998034527525e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 26200, "loss": 0.6582708470523357, "lr": 0.0003, "grad_norm": 0.13743094417910706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3833948129, "tokens_per_second": 150819.41961374847, "data_wait_s": 9.823997970670462e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 26300, "loss": 0.6972330696880817, "lr": 0.0003, "grad_norm": 0.14420341304017342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3848707744, "tokens_per_second": 149277.51460451997, "data_wait_s": 9.61299883783795e-06, "gpu_peak_gb": 85.43514251708984}
+{"step": 26400, "loss": 0.6335251517593861, "lr": 0.0003, "grad_norm": 0.1310260135444336, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3863506499, "tokens_per_second": 150632.95640157786, "data_wait_s": 1.4880999515298754e-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:bd2e51f695b76346984898a5e78f5137ac8cf22e7166fba87af5d1fedc96d600
-size 555340
+oid sha256:eac66c278e6327a96aaeaaa9a24f72bc41c17c5ddbb37fb79320c8fbbe84f501
+size 557458