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": 76945.32748866365,
-      "required_pages_per_second_upper_estimate": 0.005323256293423228
+      "required_token_positions_per_second": 76996.57474783192,
+      "required_pages_per_second_upper_estimate": 0.0053268016977225125
     },
     {
       "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": 9618.165936082956,
-      "required_pages_per_second_upper_estimate": 0.004026925227293757
+      "required_token_positions_per_second": 9624.57184347899,
+      "required_pages_per_second_upper_estimate": 0.004029607246949919
     },
     {
       "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": 19236.331872165913,
-      "required_pages_per_second_upper_estimate": 0.0040064206327864995
+      "required_token_positions_per_second": 19249.14368695798,
+      "required_pages_per_second_upper_estimate": 0.004009088995937411
     },
     {
       "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": 28854.49780824887,
-      "required_pages_per_second_upper_estimate": 0.001729783085337423
+      "required_token_positions_per_second": 28873.715530436973,
+      "required_pages_per_second_upper_estimate": 0.001730935158438837
     },
     {
       "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": 28854.49780824887,
-      "required_pages_per_second_upper_estimate": 0.0022433102870685483
+      "required_token_positions_per_second": 28873.715530436973,
+      "required_pages_per_second_upper_estimate": 0.002244804380438847
     },
     {
       "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": 28854.49780824887,
-      "required_pages_per_second_upper_estimate": 0.37011451633828285
+      "required_token_positions_per_second": 28873.715530436973,
+      "required_pages_per_second_upper_estimate": 0.3703610206441294
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.4111152180234683,
-    "producer_p95_seconds": 0.20466450000094483,
-    "gpu_update_seconds": 0.9581868374120877,
-    "producer_batches_per_second": 2.4324081331937353,
-    "gpu_batches_per_second": 1.0436377968840014,
-    "producer_headroom": 2.3307014564403454,
+    "producer_batch_seconds": 0.054008049750024156,
+    "producer_p95_seconds": 0.1821262140001636,
+    "gpu_update_seconds": 0.957549088923284,
+    "producer_batches_per_second": 18.51575838469436,
+    "gpu_batches_per_second": 1.044332882321939,
+    "producer_headroom": 17.729747571987744,
     "sustainable": true,
-    "buffer_seconds": 60.365770756961524,
+    "buffer_seconds": 60.32559260216689,
     "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": 350373.47410381865,
-    "upload_bytes_per_second": 3844602.592288011,
+    "generation_bytes_per_second": 349945.40196548984,
+    "upload_bytes_per_second": 3867724.786743171,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.1819912656395,
+    "estimated_full_shards_per_hour": 75.0901369497635,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 6021371,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.0867800260002696,
+      "bytes": 6181304,
+      "pages": 41,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -102,3 +102,4 @@
 {"step": 695200, "loss": 0.5591259002685547, "lr": 8e-05, "grad_norm": 0.19911689503341654, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101758520827, "tokens_per_second": 152081.54863085336, "data_wait_s": 1.541399979032576e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695300, "loss": 0.5476700067520142, "lr": 8e-05, "grad_norm": 0.20254950488096524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101773175001, "tokens_per_second": 152781.06473411171, "data_wait_s": 1.7416001355741173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 695400, "loss": 0.5719165205955505, "lr": 8e-05, "grad_norm": 0.2266536921798782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101787824083, "tokens_per_second": 152644.94743390344, "data_wait_s": 1.2487998901633546e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 695500, "loss": 0.5675442814826965, "lr": 8e-05, "grad_norm": 0.24226173330281586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101802556344, "tokens_per_second": 153126.22934712077, "data_wait_s": 9.756999133969657e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bf5e2963e80f8addf818363082938509ec7a367933917853f9c82ce7e5802dee
-size 222003
+oid sha256:9b47ce27e095f12fb9175f059e0a51e4a1d46055fa0c20f74eb556e160f8d3e7
+size 224121