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": 76456.20887580977,
-      "required_pages_per_second_upper_estimate": 0.005289417932874454
+      "required_token_positions_per_second": 77122.29337503343,
+      "required_pages_per_second_upper_estimate": 0.0053354992040597175
     },
     {
       "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": 9557.02610947622,
-      "required_pages_per_second_upper_estimate": 0.00400132725863828
+      "required_token_positions_per_second": 9640.28667187918,
+      "required_pages_per_second_upper_estimate": 0.004036186717438144
     },
     {
       "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": 19114.05221895244,
-      "required_pages_per_second_upper_estimate": 0.003980953005753443
+      "required_token_positions_per_second": 19280.57334375836,
+      "required_pages_per_second_upper_estimate": 0.004015634964593141
     },
     {
       "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": 28671.078328428663,
-      "required_pages_per_second_upper_estimate": 0.001718787367587531
+      "required_token_positions_per_second": 28920.86001563754,
+      "required_pages_per_second_upper_estimate": 0.0017337613983411501
     },
     {
       "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": 28671.078328428663,
-      "required_pages_per_second_upper_estimate": 0.0022290502292895563
+      "required_token_positions_per_second": 28920.86001563754,
+      "required_pages_per_second_upper_estimate": 0.002248469656796516
     },
     {
       "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": 28671.078328428663,
-      "required_pages_per_second_upper_estimate": 0.3677618081916428
+      "required_token_positions_per_second": 28920.86001563754,
+      "required_pages_per_second_upper_estimate": 0.37096573948047795
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.11135556822668491,
-    "producer_p95_seconds": 0.28215826700034086,
-    "gpu_update_seconds": 0.9643167125871847,
-    "producer_batches_per_second": 8.980242442518138,
-    "gpu_batches_per_second": 1.0370037011150413,
-    "producer_headroom": 8.659797870405,
+    "producer_batch_seconds": 0.09709914397683406,
+    "producer_p95_seconds": 0.2707410760012863,
+    "gpu_update_seconds": 0.9559881685762698,
+    "producer_batches_per_second": 10.298751966738042,
+    "gpu_batches_per_second": 1.0460380503341125,
+    "producer_headroom": 9.845485031303157,
     "sustainable": true,
-    "buffer_seconds": 60.751952892992634,
+    "buffer_seconds": 60.227254620305,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 348612.423603408,
+    "generation_bytes_per_second": 348954.90151199326,
     "upload_bytes_per_second": 4905498.855286173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.80411082340889,
+    "estimated_full_shards_per_hour": 74.87759861011361,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -263,3 +263,4 @@
 {"step": 786300, "loss": 0.6177685856819153, "lr": 3e-05, "grad_norm": 0.2900455670641432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115091916283, "tokens_per_second": 151004.5800091677, "data_wait_s": 1.8485003238311037e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786400, "loss": 0.5295373797416687, "lr": 3e-05, "grad_norm": 0.26525414383700247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115106625793, "tokens_per_second": 151712.22091129285, "data_wait_s": 1.5200999769149348e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 786500, "loss": 0.5489715337753296, "lr": 3e-05, "grad_norm": 0.256187664183011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115121285169, "tokens_per_second": 152298.42278352476, "data_wait_s": 1.4785000530537218e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 786600, "loss": 0.5553094744682312, "lr": 3e-05, "grad_norm": 0.28255089079386364, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115135798880, "tokens_per_second": 150367.19231246604, "data_wait_s": 9.563998901285231e-06, "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 sha256:97da669dfbcd12d7db52f93a339414600e762790032b8f4de2ff30335efed2cc
-size 565350
+oid sha256:92a741a9f0052e96b5aed11c4c03fd9fb816c718ab399261a3ed7b64a8c655bf
+size 567468