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": 74809.17750595936,
-      "required_pages_per_second_upper_estimate": 0.006484269904089596
+      "required_token_positions_per_second": 74838.1489943793,
+      "required_pages_per_second_upper_estimate": 0.006486781079278269
     },
     {
       "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": 9351.14718824492,
-      "required_pages_per_second_upper_estimate": 0.0037303124255006065
+      "required_token_positions_per_second": 9354.768624297412,
+      "required_pages_per_second_upper_estimate": 0.00373175707048724
     },
     {
       "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": 18702.29437648984,
-      "required_pages_per_second_upper_estimate": 0.004326446654861737
+      "required_token_positions_per_second": 18709.537248594825,
+      "required_pages_per_second_upper_estimate": 0.00432812216585303
     },
     {
       "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": 28053.441564734763,
-      "required_pages_per_second_upper_estimate": 0.0017083074062595157
+      "required_token_positions_per_second": 28064.30587289224,
+      "required_pages_per_second_upper_estimate": 0.0017089689856256863
     },
     {
       "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": 28053.441564734763,
-      "required_pages_per_second_upper_estimate": 0.0021596315885497957
+      "required_token_positions_per_second": 28064.30587289224,
+      "required_pages_per_second_upper_estimate": 0.0021604679530660884
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28053.441564734763,
-      "required_pages_per_second_upper_estimate": 0.34832985540477995
+      "required_token_positions_per_second": 28064.30587289224,
+      "required_pages_per_second_upper_estimate": 0.3484647537515954
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.27496577525010935,
-    "producer_p95_seconds": 0.26896494399989024,
-    "gpu_update_seconds": 0.9855475284984487,
-    "producer_batches_per_second": 3.6368162513694595,
-    "gpu_batches_per_second": 1.0146644084467145,
-    "producer_headroom": 3.584255268140164,
+    "producer_batch_seconds": 0.09779721589049473,
+    "producer_p95_seconds": 0.2667286840005545,
+    "gpu_update_seconds": 0.9851660014404862,
+    "producer_batches_per_second": 10.225239961020135,
+    "gpu_batches_per_second": 1.0150573594072714,
+    "producer_headroom": 10.073558766167679,
     "sustainable": true,
-    "buffer_seconds": 60.118399238405374,
+    "buffer_seconds": 60.09512608786966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 216033.8149695255,
-    "upload_bytes_per_second": 6259503.4352314435,
+    "generation_bytes_per_second": 217134.4882105055,
+    "upload_bytes_per_second": 6227574.632433899,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 46.35582768263172,
+    "estimated_full_shards_per_hour": 46.59200653778432,
     "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": 1.1122738659996685,
-      "bytes": 11663489,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.1172626269999455,
       "bytes": 11245903,
@@ -562,6 +556,12 @@
       "bytes": 5828242,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.539654905001953,
+      "bytes": 12116214,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -156,3 +156,4 @@
 {"step": 15600, "loss": 0.8388209156692028, "lr": 0.0003, "grad_norm": 0.1722140646304929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2282564906, "tokens_per_second": 150411.59326470701, "data_wait_s": 7.753000318189152e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 15700, "loss": 0.6787268780171871, "lr": 0.0003, "grad_norm": 0.1490026919468215, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2297280766, "tokens_per_second": 141902.1275588104, "data_wait_s": 1.3201002730056643e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 15800, "loss": 0.7322070933878422, "lr": 0.0003, "grad_norm": 0.17031291779696653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2311946042, "tokens_per_second": 148806.70512575624, "data_wait_s": 1.1552001524250954e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 15900, "loss": 0.6568659748882055, "lr": 0.0003, "grad_norm": 0.1520839329769262, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 2326480683, "tokens_per_second": 147825.65113271397, "data_wait_s": 1.138600055128336e-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 sha256:6c2b782c8463759e58e56f47011779b94fa90dd641c34646848804d1900b4536
-size 331559
+oid sha256:6b95fe5c34c3292f04a2ee62cdb9137253812335007ea40725c115668cbbde15
+size 333642