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": 76439.52618337888,
-      "required_pages_per_second_upper_estimate": 0.005288263785502905
+      "required_token_positions_per_second": 76646.6381172603,
+      "required_pages_per_second_upper_estimate": 0.005302592269654718
     },
     {
       "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": 9554.94077292236,
-      "required_pages_per_second_upper_estimate": 0.004000454171769958
+      "required_token_positions_per_second": 9580.829764657537,
+      "required_pages_per_second_upper_estimate": 0.004011293351985853
     },
     {
       "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": 19109.88154584472,
-      "required_pages_per_second_upper_estimate": 0.00398008436453315
+      "required_token_positions_per_second": 19161.659529315075,
+      "required_pages_per_second_upper_estimate": 0.003990868353012777
     },
     {
       "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": 28664.82231876708,
-      "required_pages_per_second_upper_estimate": 0.0017184123293606927
+      "required_token_positions_per_second": 28742.489293972616,
+      "required_pages_per_second_upper_estimate": 0.00172306834593366
     },
     {
       "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": 28664.82231876708,
-      "required_pages_per_second_upper_estimate": 0.0022285638520556468
+      "required_token_positions_per_second": 28742.489293972616,
+      "required_pages_per_second_upper_estimate": 0.0022346021177569566
     },
     {
       "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": 28664.82231876708,
-      "required_pages_per_second_upper_estimate": 0.36768156281688386
+      "required_token_positions_per_second": 28742.489293972616,
+      "required_pages_per_second_upper_estimate": 0.36867779138251966
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08722797266403859,
-    "producer_p95_seconds": 0.30199439200077904,
-    "gpu_update_seconds": 0.9645271717556972,
-    "producer_batches_per_second": 11.464212333026849,
-    "gpu_batches_per_second": 1.0367774276174435,
-    "producer_headroom": 11.05754429798117,
+    "producer_batch_seconds": 0.10589128817173332,
+    "producer_p95_seconds": 0.29155579199868953,
+    "gpu_update_seconds": 0.9619208592972449,
+    "producer_batches_per_second": 9.443647511192903,
+    "gpu_batches_per_second": 1.0395865630053753,
+    "producer_headroom": 9.084041528866967,
     "sustainable": true,
-    "buffer_seconds": 60.765211820608926,
+    "buffer_seconds": 60.60101413572643,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265602.8427663836,
-    "upload_bytes_per_second": 3206098.517676635,
+    "generation_bytes_per_second": 296268.00669930474,
+    "upload_bytes_per_second": 3163597.2924105073,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.9921871399272,
+    "estimated_full_shards_per_hour": 63.572217471450394,
     "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": 1.608142429999134,
-      "bytes": 7266564,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.734035388999473,
       "bytes": 6027691,
@@ -562,6 +556,12 @@
       "bytes": 6027005,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.4991452859976562,
+      "bytes": 1121499,
+      "pages": 36,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -318,3 +318,4 @@
 {"step": 831800, "loss": 0.5605382919311523, "lr": 3e-05, "grad_norm": 0.2931005483054918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121752186886, "tokens_per_second": 151140.40237385005, "data_wait_s": 9.47800072026439e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 831900, "loss": 0.5526161789894104, "lr": 3e-05, "grad_norm": 0.2834125292201266, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121766902416, "tokens_per_second": 151870.7840956441, "data_wait_s": 1.1004998668795452e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 832000, "loss": 0.5412599444389343, "lr": 3e-05, "grad_norm": 0.2851509734782647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121781713532, "tokens_per_second": 153708.75300884387, "data_wait_s": 7.790004019625485e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 832100, "loss": 0.5186933279037476, "lr": 3e-05, "grad_norm": 0.28658392748029904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121796373298, "tokens_per_second": 137708.14166155068, "data_wait_s": 1.0030002158600837e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b173643c7ed6f99a998cf922775716f74890b2b6f9d2b1d93cb7bbea761fc440
-size 682623
+oid sha256:dd02d4479581fc78755b9a8fef1d465087a4962b7c0131fdb6864ed909ae3100
+size 684741