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": 77361.6742888396,
-      "required_pages_per_second_upper_estimate": 0.0053520601311170726
+      "required_token_positions_per_second": 77369.02349563541,
+      "required_pages_per_second_upper_estimate": 0.005352568566295716
     },
     {
       "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": 9670.20928610495,
-      "required_pages_per_second_upper_estimate": 0.004048714691159234
+      "required_token_positions_per_second": 9671.127936954426,
+      "required_pages_per_second_upper_estimate": 0.0040490993110862995
     },
     {
       "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": 19340.4185722099,
-      "required_pages_per_second_upper_estimate": 0.004028099147454792
+      "required_token_positions_per_second": 19342.255873908853,
+      "required_pages_per_second_upper_estimate": 0.00402848180894578
     },
     {
       "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": 29010.627858314845,
-      "required_pages_per_second_upper_estimate": 0.0017391428434420954
+      "required_token_positions_per_second": 29013.38381086328,
+      "required_pages_per_second_upper_estimate": 0.0017393080585892781
     },
     {
       "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": 29010.627858314845,
-      "required_pages_per_second_upper_estimate": 0.002255448711717665
+      "required_token_positions_per_second": 29013.38381086328,
+      "required_pages_per_second_upper_estimate": 0.0022556629749061518
     },
     {
       "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": 29010.627858314845,
-      "required_pages_per_second_upper_estimate": 0.37211718498114243
+      "required_token_positions_per_second": 29013.38381086328,
+      "required_pages_per_second_upper_estimate": 0.3721525353813225
     }
   ],
   "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": 0.10314471266397618,
+    "producer_batch_seconds": 0.1005381376404273,
     "producer_p95_seconds": 0.2951992410016828,
-    "gpu_update_seconds": 0.9530300459207642,
-    "producer_batches_per_second": 9.695116445355664,
-    "gpu_batches_per_second": 1.0492848617735404,
-    "producer_headroom": 9.239737271124465,
+    "gpu_update_seconds": 0.9529395185420582,
+    "producer_batches_per_second": 9.946474278014584,
+    "gpu_batches_per_second": 1.0493845417702286,
+    "producer_headroom": 9.478388409682184,
     "sustainable": true,
-    "buffer_seconds": 60.04089289300815,
+    "buffer_seconds": 60.03518966814966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357389.0600180546,
+    "generation_bytes_per_second": 371447.7732477769,
     "upload_bytes_per_second": 5412535.765361074,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.68737268835287,
+    "estimated_full_shards_per_hour": 79.70404527735691,
     "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

@@ -293,3 +293,4 @@
 {"step": 469500, "loss": 0.5701079964637756, "lr": 0.0003, "grad_norm": 0.14553400770053482, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68720786596, "tokens_per_second": 151893.27550886673, "data_wait_s": 1.0511997970752418e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 469600, "loss": 0.5751152634620667, "lr": 0.0003, "grad_norm": 0.15159574819310465, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68735317592, "tokens_per_second": 150531.586221897, "data_wait_s": 1.3273001968627796e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 469700, "loss": 0.6009871363639832, "lr": 0.0003, "grad_norm": 0.1570330550552947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68749833614, "tokens_per_second": 151178.1979620952, "data_wait_s": 9.02800093172118e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 469800, "loss": 0.5855575203895569, "lr": 0.0003, "grad_norm": 0.1524110247525948, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68764429425, "tokens_per_second": 151972.78918288607, "data_wait_s": 1.3544002285925671e-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 sha256:3c18989e59021b97a3811bb42ccb33dda052ea5f4ad684c9a239c15ab54795f3
-size 628890
+oid sha256:91fd621abb954d2a4c6894947b247948ab61827cf8e64efd77dac6811955bbc1
+size 631008