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": 76694.17843721535,
-      "required_pages_per_second_upper_estimate": 0.005305881219297948
+      "required_token_positions_per_second": 77162.90688451091,
+      "required_pages_per_second_upper_estimate": 0.005338308940881695
     },
     {
       "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": 9586.772304651919,
-      "required_pages_per_second_upper_estimate": 0.004013781369387154
+      "required_token_positions_per_second": 9645.363360563864,
+      "required_pages_per_second_upper_estimate": 0.004038312220977107
     },
     {
       "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": 19173.544609303837,
-      "required_pages_per_second_upper_estimate": 0.003993343701743799
+      "required_token_positions_per_second": 19290.72672112773,
+      "required_pages_per_second_upper_estimate": 0.004017749645336613
     },
     {
       "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": 28760.316913955758,
-      "required_pages_per_second_upper_estimate": 0.0017241370845304498
+      "required_token_positions_per_second": 28936.09008169159,
+      "required_pages_per_second_upper_estimate": 0.001734674417546126
     },
     {
       "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": 28760.316913955758,
-      "required_pages_per_second_upper_estimate": 0.0022359881367952326
+      "required_token_positions_per_second": 28936.09008169159,
+      "required_pages_per_second_upper_estimate": 0.0022496537274422374
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28760.316913955758,
-      "required_pages_per_second_upper_estimate": 0.36890646494985646
+      "required_token_positions_per_second": 28936.09008169159,
+      "required_pages_per_second_upper_estimate": 0.3711610944150484
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09549206907814778,
+    "producer_batch_seconds": 0.12311743096074679,
     "producer_p95_seconds": 0.3376881619988126,
-    "gpu_update_seconds": 0.9613245946738503,
-    "producer_batches_per_second": 10.472073855490875,
-    "gpu_batches_per_second": 1.0402313698624044,
-    "producer_headroom": 10.067062154524391,
+    "gpu_update_seconds": 0.9554849989043064,
+    "producer_batches_per_second": 8.122326726577226,
+    "gpu_batches_per_second": 1.046588906311183,
+    "producer_headroom": 7.7607613434440585,
     "sustainable": true,
-    "buffer_seconds": 60.563449464452574,
+    "buffer_seconds": 60.195554930971305,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300692.26474825916,
-    "upload_bytes_per_second": 4448826.642554325,
+    "generation_bytes_per_second": 299306.63422947784,
+    "upload_bytes_per_second": 4490678.957494245,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.52156025729973,
+    "estimated_full_shards_per_hour": 64.22423620379688,
     "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": 1.3592714799960959,
-      "bytes": 2722837,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3202570079956786,
       "bytes": 6097373,
@@ -562,6 +556,12 @@
       "bytes": 872422,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.2673109809984453,
+      "bytes": 6309563,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -413,3 +413,4 @@
 {"step": 801300, "loss": 0.5488598346710205, "lr": 3e-05, "grad_norm": 0.2763455230250154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117286883998, "tokens_per_second": 152350.59076349984, "data_wait_s": 1.4092001947574317e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 801400, "loss": 0.5533587336540222, "lr": 3e-05, "grad_norm": 0.28289548363917666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117301622732, "tokens_per_second": 152938.3410894438, "data_wait_s": 8.970004273578525e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 801500, "loss": 0.5405556559562683, "lr": 3e-05, "grad_norm": 0.25965025620707455, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117316172213, "tokens_per_second": 151488.74701542806, "data_wait_s": 1.4527999155689031e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 801600, "loss": 0.5029669404029846, "lr": 3e-05, "grad_norm": 0.28534698327365837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117330622104, "tokens_per_second": 149872.91781004344, "data_wait_s": 1.5680998330935836e-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 sha256:60aaf8de42b3b986de3e609b59d5329fb18aef62cec72b9fcb1569f7b83df716
-size 885399
+oid sha256:535d04f6bd5bae743770af27833618db37cdacffde22c297289159e06adcc269
+size 887517