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": 77181.31413400605,
-      "required_pages_per_second_upper_estimate": 0.005339582397112472
+      "required_token_positions_per_second": 77255.46968513484,
+      "required_pages_per_second_upper_estimate": 0.005344712650204146
     },
     {
       "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": 9647.664266750757,
-      "required_pages_per_second_upper_estimate": 0.0040392755623491735
+      "required_token_positions_per_second": 9656.933710641855,
+      "required_pages_per_second_upper_estimate": 0.004043156484938377
     },
     {
       "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": 19295.328533501513,
-      "required_pages_per_second_upper_estimate": 0.00401870808149612
+      "required_token_positions_per_second": 19313.86742128371,
+      "required_pages_per_second_upper_estimate": 0.004022569242917803
     },
     {
       "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": 28942.99280025227,
-      "required_pages_per_second_upper_estimate": 0.0017350882249839979
+      "required_token_positions_per_second": 28970.801131925567,
+      "required_pages_per_second_upper_estimate": 0.0017367552920069483
     },
     {
       "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": 28942.99280025227,
-      "required_pages_per_second_upper_estimate": 0.00225019038344848
+      "required_token_positions_per_second": 28970.801131925567,
+      "required_pages_per_second_upper_estimate": 0.0022523523589201544
     },
     {
       "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": 28942.99280025227,
-      "required_pages_per_second_upper_estimate": 0.37124963507718306
+      "required_token_positions_per_second": 28970.801131925567,
+      "required_pages_per_second_upper_estimate": 0.37160633049762787
     }
   ],
   "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": 0.10943019953909072,
-    "producer_p95_seconds": 0.22018671499972697,
-    "gpu_update_seconds": 0.9552571218467435,
-    "producer_batches_per_second": 9.138245239539925,
-    "gpu_batches_per_second": 1.0468385706109762,
-    "producer_headroom": 8.729373846252615,
+    "producer_batch_seconds": 0.06196427735150678,
+    "producer_p95_seconds": 0.21353033499872254,
+    "gpu_update_seconds": 0.9543401949465646,
+    "producer_batches_per_second": 16.138330708309038,
+    "gpu_batches_per_second": 1.0478443696442985,
+    "producer_headroom": 15.401457674279776,
     "sustainable": true,
-    "buffer_seconds": 60.18119867634484,
+    "buffer_seconds": 60.12343228163357,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 359418.9228176706,
+    "generation_bytes_per_second": 345586.2764468814,
     "upload_bytes_per_second": 5977036.19421587,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.12293399236287,
+    "estimated_full_shards_per_hour": 74.15477008871872,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -137,3 +137,4 @@
 {"step": 498700, "loss": 0.584259569644928, "lr": 0.0003, "grad_norm": 0.1385729513594764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72992965890, "tokens_per_second": 152308.28385024078, "data_wait_s": 1.1232999895582907e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 498800, "loss": 0.5576951503753662, "lr": 0.0003, "grad_norm": 0.1574530003632909, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73007594670, "tokens_per_second": 152748.16180447643, "data_wait_s": 1.1441999959060922e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 498900, "loss": 0.6467173099517822, "lr": 0.0003, "grad_norm": 0.16527652480198224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73022166926, "tokens_per_second": 152253.54614836109, "data_wait_s": 1.338299989583902e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 499000, "loss": 0.6090105772018433, "lr": 0.0003, "grad_norm": 0.16052634128681137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73036684139, "tokens_per_second": 151162.77354802587, "data_wait_s": 1.4625999028794467e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:099a8e13c41c4367719aebee158179dd76a8d62f4c0a377728003629bf83d4bf
-size 296133
+oid sha256:c39aad1cfdc9612eb756fded259d055f018f881ed3621384bf37e99d0620ec76
+size 298251