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": 77001.63110820118,
-      "required_pages_per_second_upper_estimate": 0.0053271515084652304
+      "required_token_positions_per_second": 76339.08332236053,
+      "required_pages_per_second_upper_estimate": 0.005281314915318109
     },
     {
       "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": 9625.203888525148,
-      "required_pages_per_second_upper_estimate": 0.0040298718710121435
+      "required_token_positions_per_second": 9542.385415295066,
+      "required_pages_per_second_upper_estimate": 0.003995197505717091
     },
     {
       "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": 19250.407777050295,
-      "required_pages_per_second_upper_estimate": 0.004009352272567342
+      "required_token_positions_per_second": 19084.770830590132,
+      "required_pages_per_second_upper_estimate": 0.003974854464759713
     },
     {
       "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": 28875.61166557544,
-      "required_pages_per_second_upper_estimate": 0.0017310488288451574
+      "required_token_positions_per_second": 28627.1562458852,
+      "required_pages_per_second_upper_estimate": 0.001716154305804187
     },
     {
       "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": 28875.61166557544,
-      "required_pages_per_second_upper_estimate": 0.002244951796605646
+      "required_token_positions_per_second": 28627.1562458852,
+      "required_pages_per_second_upper_estimate": 0.002225635480564626
     },
     {
       "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": 28875.61166557544,
-      "required_pages_per_second_upper_estimate": 0.3703853422297744
+      "required_token_positions_per_second": 28627.1562458852,
+      "required_pages_per_second_upper_estimate": 0.3671984228766332
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.36521766404649725,
-    "producer_p95_seconds": 0.3560835930038593,
-    "gpu_update_seconds": 0.9574862108621941,
-    "producer_batches_per_second": 2.738093193303723,
-    "gpu_batches_per_second": 1.0444014635986487,
-    "producer_headroom": 2.621686476643947,
+    "producer_batch_seconds": 0.13003157482808092,
+    "producer_p95_seconds": 0.3643983629954164,
+    "gpu_update_seconds": 0.9657962447448499,
+    "producer_batches_per_second": 7.690439812961839,
+    "gpu_batches_per_second": 1.0354150841248986,
+    "producer_headroom": 7.42739789179483,
     "sustainable": true,
-    "buffer_seconds": 60.32163128431823,
+    "buffer_seconds": 60.845163418925544,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 284724.7621214127,
+    "generation_bytes_per_second": 285322.84491709917,
     "upload_bytes_per_second": 5382416.4966704035,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.095305898015845,
+    "estimated_full_shards_per_hour": 61.223640543315234,
     "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": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -415,3 +415,4 @@
 {"step": 481700, "loss": 0.5596342086791992, "lr": 0.0003, "grad_norm": 0.1570385359003396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70504874513, "tokens_per_second": 151225.64996765272, "data_wait_s": 1.6231999325100332e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 481800, "loss": 0.5760315656661987, "lr": 0.0003, "grad_norm": 0.16037769213624076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70519411476, "tokens_per_second": 150152.869500968, "data_wait_s": 1.2286000128369778e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 481900, "loss": 0.63841313123703, "lr": 0.0003, "grad_norm": 0.19331604354307996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70533946419, "tokens_per_second": 150540.1852756703, "data_wait_s": 1.0085997928399593e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 482000, "loss": 0.6147832274436951, "lr": 0.0003, "grad_norm": 0.1746085706456048, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70548781533, "tokens_per_second": 153890.00997796978, "data_wait_s": 1.3793993275612593e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dced5d9385ba9e0ffd60c97e194885f84cba363a28ce0e618d85f0910adf63f6
-size 889635
+oid sha256:6b229fa952e50af2ff730a7c6158b8ee7c974a2cfb263c0ad66aeb8fa1e95b01
+size 891753