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": 77443.0060429489,
-      "required_pages_per_second_upper_estimate": 0.005357686850582024
+      "required_token_positions_per_second": 77621.13671610574,
+      "required_pages_per_second_upper_estimate": 0.0053700103438711205
     },
     {
       "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": 9680.375755368612,
-      "required_pages_per_second_upper_estimate": 0.00405297117953991
+      "required_token_positions_per_second": 9702.642089513218,
+      "required_pages_per_second_upper_estimate": 0.004062293628672326
     },
     {
       "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": 19360.751510737224,
-      "required_pages_per_second_upper_estimate": 0.004032333962334386
+      "required_token_positions_per_second": 19405.284179026436,
+      "required_pages_per_second_upper_estimate": 0.00404160894273359
     },
     {
       "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": 29041.127266105836,
-      "required_pages_per_second_upper_estimate": 0.001740971236369266
+      "required_token_positions_per_second": 29107.92626853965,
+      "required_pages_per_second_upper_estimate": 0.0017449757345690022
     },
     {
       "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": 29041.127266105836,
-      "required_pages_per_second_upper_estimate": 0.0022578199065206974
+      "required_token_positions_per_second": 29107.92626853965,
+      "required_pages_per_second_upper_estimate": 0.0022630132351421665
     },
     {
       "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": 29041.127266105836,
-      "required_pages_per_second_upper_estimate": 0.3725083986365726
+      "required_token_positions_per_second": 29107.92626853965,
+      "required_pages_per_second_upper_estimate": 0.3733652245166128
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07517217553903777,
-    "producer_p95_seconds": 0.20055416399918613,
-    "gpu_update_seconds": 0.9520291601169433,
-    "producer_batches_per_second": 13.302794455917915,
-    "gpu_batches_per_second": 1.050387994289129,
-    "producer_headroom": 12.664648233075862,
+    "producer_batch_seconds": 0.05838737730468324,
+    "producer_p95_seconds": 0.19753887000115355,
+    "gpu_update_seconds": 0.94984437382894,
+    "producer_batches_per_second": 17.12698953374277,
+    "gpu_batches_per_second": 1.052804046171139,
+    "producer_headroom": 16.26797464925271,
     "sustainable": true,
-    "buffer_seconds": 60.92986624748437,
+    "buffer_seconds": 60.79003992505216,
     "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": 516854.66904558876,
+    "generation_bytes_per_second": 517928.6847646722,
     "upload_bytes_per_second": 5802177.8237436805,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.90498021627185,
+    "estimated_full_shards_per_hour": 111.1354389877808,
     "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

@@ -99,3 +99,4 @@
 {"step": 494900, "loss": 0.5454460978507996, "lr": 0.0003, "grad_norm": 0.15774258474687167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72437579468, "tokens_per_second": 152201.55731662142, "data_wait_s": 2.1256000763969496e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 495000, "loss": 0.5722364187240601, "lr": 0.0003, "grad_norm": 0.1551653797292115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72452172927, "tokens_per_second": 152190.61102034835, "data_wait_s": 1.1348000043653883e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 495100, "loss": 0.5845643281936646, "lr": 0.0003, "grad_norm": 0.14639399152855526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72466684245, "tokens_per_second": 135075.90875019145, "data_wait_s": 1.6693998986738734e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 495200, "loss": 0.5805988311767578, "lr": 0.0003, "grad_norm": 0.15173483243161145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72481139885, "tokens_per_second": 151140.1953230263, "data_wait_s": 1.5265999536495656e-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 sha256:7f36f227f118b3c77b73b49807d2a6977cca31e60a416a39bb788f579383900f
-size 215649
+oid sha256:6be829f95ee8dede7f07b2d9a55174294b6870ea1f5a81241facd7d0ae21c0c8
+size 217767