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": 77235.25295698272,
-      "required_pages_per_second_upper_estimate": 0.00534331401004131
+      "required_token_positions_per_second": 77426.30200708305,
+      "required_pages_per_second_upper_estimate": 0.0053565312266221216
     },
     {
       "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": 9654.40661962284,
-      "required_pages_per_second_upper_estimate": 0.004042098444700376
+      "required_token_positions_per_second": 9678.287750885382,
+      "required_pages_per_second_upper_estimate": 0.004052096975665273
     },
     {
       "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": 19308.81323924568,
-      "required_pages_per_second_upper_estimate": 0.0040215165900870245
+      "required_token_positions_per_second": 19356.575501770763,
+      "required_pages_per_second_upper_estimate": 0.004031464209795435
     },
     {
       "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": 28963.21985886852,
-      "required_pages_per_second_upper_estimate": 0.001736300806263112
+      "required_token_positions_per_second": 29034.863252656145,
+      "required_pages_per_second_upper_estimate": 0.001740595718327552
     },
     {
       "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": 28963.21985886852,
-      "required_pages_per_second_upper_estimate": 0.002251762948286465
+      "required_token_positions_per_second": 29034.863252656145,
+      "required_pages_per_second_upper_estimate": 0.0022573329070274674
     },
     {
       "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": 28963.21985886852,
-      "required_pages_per_second_upper_estimate": 0.3715090860669889
+      "required_token_positions_per_second": 29034.863252656145,
+      "required_pages_per_second_upper_estimate": 0.3724280505978136
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08726010233601755,
-    "producer_p95_seconds": 0.2680390780005837,
-    "gpu_update_seconds": 0.9545899984436365,
-    "producer_batches_per_second": 11.459991144053921,
-    "gpu_batches_per_second": 1.0475701627194922,
-    "producer_headroom": 10.939592928366519,
+    "producer_batch_seconds": 0.07070974933591856,
+    "producer_p95_seconds": 0.25876564800273627,
+    "gpu_update_seconds": 0.9522345519388912,
+    "producer_batches_per_second": 14.142321382718128,
+    "gpu_batches_per_second": 1.0501614313026673,
+    "producer_headroom": 13.466807065248396,
     "sustainable": true,
-    "buffer_seconds": 60.139169901949096,
+    "buffer_seconds": 60.943011324089035,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340755.37563627755,
+    "generation_bytes_per_second": 386715.169911751,
     "upload_bytes_per_second": 5585087.947976237,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.11817123237843,
+    "estimated_full_shards_per_hour": 82.9800731946411,
     "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

@@ -194,3 +194,4 @@
 {"step": 504400, "loss": 0.5893999338150024, "lr": 0.0003, "grad_norm": 0.15345058368615985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73827340094, "tokens_per_second": 150455.18426398194, "data_wait_s": 1.47380014823284e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 504500, "loss": 0.5588216185569763, "lr": 0.0003, "grad_norm": 0.14769947251139195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73842026382, "tokens_per_second": 153087.75038712792, "data_wait_s": 1.0253999789711088e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 504600, "loss": 0.5975275635719299, "lr": 0.0003, "grad_norm": 0.1692339807413062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73856723021, "tokens_per_second": 153526.31281645794, "data_wait_s": 1.3040000339969993e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 504700, "loss": 0.5502405166625977, "lr": 0.0003, "grad_norm": 0.16135350636367934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73871415633, "tokens_per_second": 153141.15619234374, "data_wait_s": 1.1763000657083467e-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 sha256:23f037460c93e26580d9512f06c0d004e63c839ef6c13f677e3598718fdcae47
-size 417642
+oid sha256:14346b8d31b44ea4444be3ab90ad20723209be0471589fdfdc7defc2a45fe706
+size 419760