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": 77294.85828530704,
-      "required_pages_per_second_upper_estimate": 0.005347437644958203
+      "required_token_positions_per_second": 77284.0585701424,
+      "required_pages_per_second_upper_estimate": 0.005346690495604321
     },
     {
       "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": 9661.85728566338,
-      "required_pages_per_second_upper_estimate": 0.004045217882983951
+      "required_token_positions_per_second": 9660.5073212678,
+      "required_pages_per_second_upper_estimate": 0.004044652681081984
     },
     {
       "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": 19323.71457132676,
-      "required_pages_per_second_upper_estimate": 0.004024620144584961
+      "required_token_positions_per_second": 19321.0146425356,
+      "required_pages_per_second_upper_estimate": 0.004024057820619672
     },
     {
       "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": 28985.57185699014,
-      "required_pages_per_second_upper_estimate": 0.001737640774420978
+      "required_token_positions_per_second": 28981.5219638034,
+      "required_pages_per_second_upper_estimate": 0.001737397989508262
     },
     {
       "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": 28985.57185699014,
-      "required_pages_per_second_upper_estimate": 0.002253500717824372
+      "required_token_positions_per_second": 28981.5219638034,
+      "required_pages_per_second_upper_estimate": 0.0022531858564427
     },
     {
       "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": 28985.57185699014,
-      "required_pages_per_second_upper_estimate": 0.3717957934991873
+      "required_token_positions_per_second": 28981.5219638034,
+      "required_pages_per_second_upper_estimate": 0.37174384581782427
     }
   ],
   "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": 0.2252796677029778,
-    "producer_p95_seconds": 0.3136702810006682,
-    "gpu_update_seconds": 0.9538538737966085,
-    "producer_batches_per_second": 4.438927002140557,
-    "gpu_batches_per_second": 1.0483786117256273,
-    "producer_headroom": 4.2340877164921356,
+    "producer_batch_seconds": 0.1143439260237642,
+    "producer_p95_seconds": 0.30335024499800056,
+    "gpu_update_seconds": 0.9539871658407413,
+    "producer_batches_per_second": 8.745545432751445,
+    "gpu_batches_per_second": 1.048232131213954,
+    "producer_headroom": 8.34313810112199,
     "sustainable": true,
-    "buffer_seconds": 60.09279404918633,
+    "buffer_seconds": 60.1011914479667,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 532950.9132587063,
+    "generation_bytes_per_second": 531159.2492183896,
     "upload_bytes_per_second": 5857714.877073814,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.35885952301877,
+    "estimated_full_shards_per_hour": 113.97441012777107,
     "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

@@ -318,3 +318,4 @@
 {"step": 516800, "loss": 0.5631062984466553, "lr": 0.0003, "grad_norm": 0.155585051464483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75641993544, "tokens_per_second": 152525.13666541467, "data_wait_s": 1.7769001715350896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516900, "loss": 0.5886542797088623, "lr": 0.0003, "grad_norm": 0.1478557523350261, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75656572274, "tokens_per_second": 151946.24390414, "data_wait_s": 2.0961000700481236e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 517000, "loss": 0.6103763580322266, "lr": 0.0003, "grad_norm": 0.17606200312496753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75671196514, "tokens_per_second": 152217.15486600713, "data_wait_s": 1.4236000424716622e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 517100, "loss": 0.6098175644874573, "lr": 0.0003, "grad_norm": 0.1698224621202967, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75685702108, "tokens_per_second": 151532.1575203471, "data_wait_s": 1.554000482428819e-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 sha256:564b8e5715f9c9f79aa59dd0da451fd3d0d1b048d888261d65164550a2b9567c
-size 682623
+oid sha256:b59ca0b1e00f422b559b4ccb41320c74b92c6144e1517c6c58975678198ab200
+size 684741