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": 75609.17320586169,
-      "required_pages_per_second_upper_estimate": 0.005230818039833913
+      "required_token_positions_per_second": 76271.55794822388,
+      "required_pages_per_second_upper_estimate": 0.005276643353254882
     },
     {
       "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": 9451.146650732711,
-      "required_pages_per_second_upper_estimate": 0.003956997740276894
+      "required_token_positions_per_second": 9533.944743527985,
+      "required_pages_per_second_upper_estimate": 0.003991663572709484
     },
     {
       "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": 18902.293301465423,
-      "required_pages_per_second_upper_estimate": 0.003936849207699089
+      "required_token_positions_per_second": 19067.88948705597,
+      "required_pages_per_second_upper_estimate": 0.003971338526092514
     },
     {
       "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": 28353.439952198132,
-      "required_pages_per_second_upper_estimate": 0.0016997454319382284
+      "required_token_positions_per_second": 28601.834230583954,
+      "required_pages_per_second_upper_estimate": 0.0017146362896513561
     },
     {
       "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": 28353.439952198132,
-      "required_pages_per_second_upper_estimate": 0.002204355242681195
+      "required_token_positions_per_second": 28601.834230583954,
+      "required_pages_per_second_upper_estimate": 0.0022236668052547284
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28353.439952198132,
-      "required_pages_per_second_upper_estimate": 0.363687484154874
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28601.834230583954,
+      "required_pages_per_second_upper_estimate": 0.36687361925301054
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.37327320978134537,
-    "producer_p95_seconds": 0.24072316799902183,
-    "gpu_update_seconds": 0.9751197754703679,
-    "producer_batches_per_second": 2.679002869200756,
-    "gpu_batches_per_second": 1.0255150445673513,
-    "producer_headroom": 2.6123486762995127,
+    "producer_batch_seconds": 0.25474713546094563,
+    "producer_p95_seconds": 0.2564348660016549,
+    "gpu_update_seconds": 0.9666512915607343,
+    "producer_batches_per_second": 3.9254612154542023,
+    "gpu_batches_per_second": 1.0344992126223942,
+    "producer_headroom": 3.7945521538903746,
     "sustainable": true,
-    "buffer_seconds": 60.45742607916281,
+    "buffer_seconds": 60.899031368326256,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 398124.5243090257,
-    "upload_bytes_per_second": 4918447.240584663,
+    "generation_bytes_per_second": 381562.12851987366,
+    "upload_bytes_per_second": 4993161.180169942,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.42825505211906,
+    "estimated_full_shards_per_hour": 81.87435046860845,
     "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": [
-    {
-      "seconds": 1.0340174030006892,
-      "bytes": 2678432,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1785862379992977,
       "bytes": 9722174,
@@ -562,6 +556,12 @@
       "bytes": 2603502,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.5727497110001423,
+      "bytes": 11757823,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 732500, "loss": 0.5607808828353882, "lr": 5e-05, "grad_norm": 0.2701238554798232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107217652722, "tokens_per_second": 153406.1708068083, "data_wait_s": 1.7020000086631626e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732600, "loss": 0.5459712743759155, "lr": 5e-05, "grad_norm": 0.22150784871969925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107232409461, "tokens_per_second": 150610.48302994028, "data_wait_s": 1.8864000594476238e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 732700, "loss": 0.5102741122245789, "lr": 5e-05, "grad_norm": 0.21796404490415877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107247016435, "tokens_per_second": 149856.62709858778, "data_wait_s": 1.8905999240814708e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 732800, "loss": 0.6318305134773254, "lr": 5e-05, "grad_norm": 0.2544219996035508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107261768006, "tokens_per_second": 150330.6799940051, "data_wait_s": 1.886000063677784e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b2312787e030b7dd611474f85055e2ed781405eab7f6749b5c537aa91b8733c9
-size 270717
+oid sha256:48e1b11c73e108f3a05baf3650e02b6b984fc823ae9f7c9ec7a21044fcf27f32
+size 272835