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": 75745.19625580676,
-      "required_pages_per_second_upper_estimate": 0.005240228430046073
+      "required_token_positions_per_second": 76142.90764269666,
+      "required_pages_per_second_upper_estimate": 0.005267743026608674
     },
     {
       "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": 9468.149531975845,
-      "required_pages_per_second_upper_estimate": 0.003964116491592859
+      "required_token_positions_per_second": 9517.863455337083,
+      "required_pages_per_second_upper_estimate": 0.003984930673159437
     },
     {
       "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": 18936.29906395169,
-      "required_pages_per_second_upper_estimate": 0.003943931711232715
+      "required_token_positions_per_second": 19035.726910674166,
+      "required_pages_per_second_upper_estimate": 0.003964639909616361
     },
     {
       "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": 28404.448595927533,
-      "required_pages_per_second_upper_estimate": 0.0017028033222441167
+      "required_token_positions_per_second": 28553.590366011253,
+      "required_pages_per_second_upper_estimate": 0.0017117441436343387
     },
     {
       "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": 28404.448595927533,
-      "required_pages_per_second_upper_estimate": 0.0022083209403678437
+      "required_token_positions_per_second": 28553.590366011253,
+      "required_pages_per_second_upper_estimate": 0.002219916056986535
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28404.448595927533,
-      "required_pages_per_second_upper_estimate": 0.3643417682678202
+      "required_token_positions_per_second": 28553.590366011253,
+      "required_pages_per_second_upper_estimate": 0.36625479875849787
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.16442591646878846,
-    "producer_p95_seconds": 0.13880198200013183,
-    "gpu_update_seconds": 0.973368657611048,
-    "producer_batches_per_second": 6.081766314435117,
-    "gpu_batches_per_second": 1.0273599752577955,
-    "producer_headroom": 5.919800713385801,
+    "producer_batch_seconds": 0.06299420892969465,
+    "producer_p95_seconds": 0.13468704099977913,
+    "gpu_update_seconds": 0.9682845360459741,
+    "producer_batches_per_second": 15.87447508255974,
+    "gpu_batches_per_second": 1.032754281178069,
+    "producer_headroom": 15.371008740289733,
     "sustainable": true,
-    "buffer_seconds": 60.348856771884975,
+    "buffer_seconds": 60.03364123485039,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 69,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424533.6472953534,
-    "upload_bytes_per_second": 5696740.445231295,
+    "generation_bytes_per_second": 443001.95457993686,
+    "upload_bytes_per_second": 5782868.745466187,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.09503807206585,
+    "estimated_full_shards_per_hour": 95.05790689514713,
     "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": [
@@ -250,6 +250,18 @@
       "bytes": 7640700,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.6995113520001723,
+      "bytes": 11765053,
+      "pages": 37,
+      "shards": 3
+    },
+    {
+      "seconds": 1.1097027079999862,
+      "bytes": 5584212,
+      "pages": 18,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790911926322536473-f9bff804/train.jsonl

@@ -16,3 +16,4 @@
 {"step": 484800, "loss": 0.5741428136825562, "lr": 0.0003, "grad_norm": 0.13144694445292182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70958166119, "tokens_per_second": 152415.5815909531, "data_wait_s": 1.4009000096848467e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 484900, "loss": 0.5589832663536072, "lr": 0.0003, "grad_norm": 0.17102171825404996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70972921897, "tokens_per_second": 152395.1114727197, "data_wait_s": 9.685999884823104e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 485000, "loss": 0.5554909706115723, "lr": 0.0003, "grad_norm": 0.1641477165417348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 70987550848, "tokens_per_second": 152536.30470855438, "data_wait_s": 9.10300013856613e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 485100, "loss": 0.5988941788673401, "lr": 0.0003, "grad_norm": 0.17389446853095353, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71002266879, "tokens_per_second": 139459.88657819288, "data_wait_s": 9.113000032812124e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790911926.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.444.0.1790911926322536473-f9bff804

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f1b8a19c4cd3cd25a9b2e1f7530e8e5734e55706d26f801384c650eb825a45c4
-size 39072
+oid sha256:1efaf453f47215f67ff5033238c38888d2b8580f9b31b91a24c409f1debe571e
+size 41190