Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76269.8945027955,
-      "required_pages_per_second_upper_estimate": 0.005276528272240421
+      "required_token_positions_per_second": 76171.99284248512,
+      "required_pages_per_second_upper_estimate": 0.005269755208217
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9533.736812849438,
-      "required_pages_per_second_upper_estimate": 0.0039915765164764624
+      "required_token_positions_per_second": 9521.49910531064,
+      "required_pages_per_second_upper_estimate": 0.003986452843882361
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19067.473625698876,
-      "required_pages_per_second_upper_estimate": 0.003971251913138833
+      "required_token_positions_per_second": 19042.99821062128,
+      "required_pages_per_second_upper_estimate": 0.00396615432963827
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28601.21043854831,
-      "required_pages_per_second_upper_estimate": 0.0017145988942712945
+      "required_token_positions_per_second": 28564.497315931923,
+      "required_pages_per_second_upper_estimate": 0.0017123979986281344
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28601.21043854831,
-      "required_pages_per_second_upper_estimate": 0.0022236183081677284
+      "required_token_positions_per_second": 28564.497315931923,
+      "required_pages_per_second_upper_estimate": 0.0022207640243682644
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28601.21043854831,
-      "required_pages_per_second_upper_estimate": 0.366865617918553
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28564.497315931923,
+      "required_pages_per_second_upper_estimate": 0.3663947014011098
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09373406412487384,
-    "producer_p95_seconds": 0.27547786699869903,
-    "gpu_update_seconds": 0.9666723742131001,
-    "producer_batches_per_second": 10.668480123381675,
-    "gpu_batches_per_second": 1.0344766506998087,
-    "producer_headroom": 10.312925010114633,
+    "producer_batch_seconds": 0.07654892551559556,
+    "producer_p95_seconds": 0.2765441150004335,
+    "gpu_update_seconds": 0.9679148102697139,
+    "producer_batches_per_second": 13.063540647559668,
+    "gpu_batches_per_second": 1.0331487744477692,
+    "producer_headroom": 12.644394467333413,
     "sustainable": true,
-    "buffer_seconds": 60.90035957542531,
+    "buffer_seconds": 60.01071823672226,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371315.40326317993,
-    "upload_bytes_per_second": 4874696.5722419,
+    "generation_bytes_per_second": 372253.45911871217,
+    "upload_bytes_per_second": 4923364.359075203,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.67564176007794,
+    "estimated_full_shards_per_hour": 79.87692670985244,
     "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": 1.9888146960001905,
-      "bytes": 6170248,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.596614274996682,
       "bytes": 11247081,
@@ -562,6 +556,12 @@
       "bytes": 7625416,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.1141033280000556,
+      "bytes": 6156266,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -290,3 +290,4 @@
 {"step": 789000, "loss": 0.5050582885742188, "lr": 3e-05, "grad_norm": 0.2650915935230674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115486997166, "tokens_per_second": 151954.43834289283, "data_wait_s": 9.687999408924952e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 789100, "loss": 0.45617660880088806, "lr": 3e-05, "grad_norm": 0.2658275310719928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115501707800, "tokens_per_second": 152874.7725302047, "data_wait_s": 8.409002475673333e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 789200, "loss": 0.5413856506347656, "lr": 3e-05, "grad_norm": 0.272991260775642, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115516445817, "tokens_per_second": 152943.70408638308, "data_wait_s": 1.0864998330362141e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 789300, "loss": 0.5332176089286804, "lr": 3e-05, "grad_norm": 0.26102109338624185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115530997889, "tokens_per_second": 151776.20846422215, "data_wait_s": 1.37959978019353e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee935c783824bc780490a2d7c7a1082617a7aad09abedf56507d089f2bf3786c
-size 622536
+oid sha256:8541ac05191368ade794e4da0d7e2f6ee7cf8eaf677dfb483fc3190890b164f1
+size 624654