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": 76067.52949096054,
-      "required_pages_per_second_upper_estimate": 0.00659334066488555
+      "required_token_positions_per_second": 75910.38584387765,
+      "required_pages_per_second_upper_estimate": 0.006579719851833337
     },
     {
       "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": 9508.441186370068,
-      "required_pages_per_second_upper_estimate": 0.0037930593531075745
+      "required_token_positions_per_second": 9488.798230484706,
+      "required_pages_per_second_upper_estimate": 0.0037852234843165416
     },
     {
       "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": 19016.882372740136,
-      "required_pages_per_second_upper_estimate": 0.004399221051234606
+      "required_token_positions_per_second": 18977.596460969413,
+      "required_pages_per_second_upper_estimate": 0.004390132946954882
     },
     {
       "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": 28525.323559110202,
-      "required_pages_per_second_upper_estimate": 0.0017370425439434932
+      "required_token_positions_per_second": 28466.39469145412,
+      "required_pages_per_second_upper_estimate": 0.0017334540850791119
     },
     {
       "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": 28525.323559110202,
-      "required_pages_per_second_upper_estimate": 0.0021959583707297815
+      "required_token_positions_per_second": 28466.39469145412,
+      "required_pages_per_second_upper_estimate": 0.00219142186337207
     },
     {
       "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": 28525.323559110202,
-      "required_pages_per_second_upper_estimate": 0.35418905048748034
+      "required_token_positions_per_second": 28466.39469145412,
+      "required_pages_per_second_upper_estimate": 0.35345735117342486
     }
   ],
   "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": 0.0993628385701868,
-    "producer_p95_seconds": 0.3546145770014846,
-    "gpu_update_seconds": 0.9692440453026865,
-    "producer_batches_per_second": 10.06412472097032,
-    "gpu_batches_per_second": 1.0317318995627243,
-    "producer_headroom": 9.754592956984043,
+    "producer_batch_seconds": 0.14032737023438813,
+    "producer_p95_seconds": 0.3465929919984774,
+    "gpu_update_seconds": 0.9712504972854954,
+    "producer_batches_per_second": 7.126193545348316,
+    "gpu_batches_per_second": 1.0296005024397468,
+    "producer_headroom": 6.921319024672239,
     "sustainable": true,
-    "buffer_seconds": 60.09313080876656,
+    "buffer_seconds": 60.217530831700714,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344124.8088828635,
-    "upload_bytes_per_second": 4891827.717010798,
+    "generation_bytes_per_second": 373695.8832727791,
+    "upload_bytes_per_second": 4833210.72004389,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.84117317070417,
+    "estimated_full_shards_per_hour": 80.18643735539942,
     "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": 0.9954401819995837,
-      "bytes": 6010708,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.1992983269992692,
       "bytes": 13571269,
@@ -562,6 +556,12 @@
       "bytes": 6277468,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.9982111080025788,
+      "bytes": 877422,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -388,3 +388,4 @@
 {"step": 193800, "loss": 0.5223269741982222, "lr": 0.0003, "grad_norm": 0.21479016209627066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28364727251, "tokens_per_second": 137657.85232819806, "data_wait_s": 1.0907002433668822e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 193900, "loss": 0.5552724972367287, "lr": 0.0003, "grad_norm": 0.13167411959697956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28379301141, "tokens_per_second": 149164.66600129788, "data_wait_s": 8.495000656694174e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 194000, "loss": 0.7089795097708702, "lr": 0.0003, "grad_norm": 0.18382650708824108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28393808966, "tokens_per_second": 149189.65971278542, "data_wait_s": 9.483002941124141e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 194100, "loss": 0.5605682022869587, "lr": 0.0003, "grad_norm": 0.1346691769896389, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28408327117, "tokens_per_second": 149352.59890534455, "data_wait_s": 8.06100433692336e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:363bc7bc5e6975318ce2e42982abf185e27cd0ded02849794038697820830af9
-size 831666
+oid sha256:25c11407209acf313c6338ae993064266550c532c23fdaecc84a134d8a4038f0
+size 833784