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": 77261.44503327008,
-      "required_pages_per_second_upper_estimate": 0.00534512603865286
+      "required_token_positions_per_second": 77233.90837115186,
+      "required_pages_per_second_upper_estimate": 0.005343220988473662
     },
     {
       "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": 9657.68062915876,
-      "required_pages_per_second_upper_estimate": 0.004043469204123972
+      "required_token_positions_per_second": 9654.238546393983,
+      "required_pages_per_second_upper_estimate": 0.004042028075949222
     },
     {
       "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": 19315.36125831752,
-      "required_pages_per_second_upper_estimate": 0.004022880369776814
+      "required_token_positions_per_second": 19308.477092787965,
+      "required_pages_per_second_upper_estimate": 0.004021446579644661
     },
     {
       "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": 28973.041887476276,
-      "required_pages_per_second_upper_estimate": 0.0017368896218807777
+      "required_token_positions_per_second": 28962.715639181948,
+      "required_pages_per_second_upper_estimate": 0.0017362705790628008
     },
     {
       "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": 28973.041887476276,
-      "required_pages_per_second_upper_estimate": 0.00225252656780818
+      "required_token_positions_per_second": 28962.715639181948,
+      "required_pages_per_second_upper_estimate": 0.002251723747423662
     },
     {
       "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": 28973.041887476276,
-      "required_pages_per_second_upper_estimate": 0.3716350725038965
+      "required_token_positions_per_second": 28962.715639181948,
+      "required_pages_per_second_upper_estimate": 0.37150261847823846
     }
   ],
   "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": 0.12061048737535884,
-    "producer_p95_seconds": 0.3548598529960145,
-    "gpu_update_seconds": 0.9542663869184881,
-    "producer_batches_per_second": 8.291152964898005,
-    "gpu_batches_per_second": 1.0479254154903168,
-    "producer_headroom": 7.91196858320173,
+    "producer_batch_seconds": 0.1012015116327234,
+    "producer_p95_seconds": 0.36666951500228606,
+    "gpu_update_seconds": 0.9546066171570132,
+    "producer_batches_per_second": 9.881275327478914,
+    "gpu_batches_per_second": 1.0475519256069858,
+    "producer_headroom": 9.432730813561705,
     "sustainable": true,
-    "buffer_seconds": 60.11878237586475,
+    "buffer_seconds": 60.140216880891835,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420290.59650202515,
+    "generation_bytes_per_second": 436605.67077275255,
     "upload_bytes_per_second": 5302101.0399973355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.18457814498487,
+    "estimated_full_shards_per_hour": 93.68541328799184,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -430,3 +430,4 @@
 {"step": 600100, "loss": 0.6710374355316162, "lr": 0.0003, "grad_norm": 0.1682189764871132, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87836984173, "tokens_per_second": 138292.41131993674, "data_wait_s": 1.3665005099028349e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600200, "loss": 0.5465240478515625, "lr": 0.0003, "grad_norm": 0.1641682083351951, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87851654044, "tokens_per_second": 153067.15591660165, "data_wait_s": 1.2889002391602844e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 600300, "loss": 0.5746079087257385, "lr": 0.0003, "grad_norm": 0.16822399315171796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87866176121, "tokens_per_second": 150920.4173547582, "data_wait_s": 9.325005521532148e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 600400, "loss": 0.5703005790710449, "lr": 0.0003, "grad_norm": 0.16658278122116918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87880667765, "tokens_per_second": 150292.4256544484, "data_wait_s": 1.6500001947861165e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f277d77adf36de633555645436ee466166cd68e02c48213dafe314fa74c6f0c
-size 922188
+oid sha256:328bdda752da46c725d39c9a376072b592a6cb4c339864aa273b2d3eea82da88
+size 924306