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": 76347.73214395717,
-      "required_pages_per_second_upper_estimate": 0.005281913260864196
+      "required_token_positions_per_second": 76630.97624320345,
+      "required_pages_per_second_upper_estimate": 0.00530150874486691
     },
     {
       "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": 9543.466517994646,
-      "required_pages_per_second_upper_estimate": 0.003995650140841414
+      "required_token_positions_per_second": 9578.872030400431,
+      "required_pages_per_second_upper_estimate": 0.004010473689534542
     },
     {
       "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": 19086.933035989292,
-      "required_pages_per_second_upper_estimate": 0.003975304795123167
+      "required_token_positions_per_second": 19157.744060800862,
+      "required_pages_per_second_upper_estimate": 0.0039900528641791146
     },
     {
       "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": 28630.399553983938,
-      "required_pages_per_second_upper_estimate": 0.0017163487371724619
+      "required_token_positions_per_second": 28736.616091201293,
+      "required_pages_per_second_upper_estimate": 0.0017227162563953806
     },
     {
       "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": 28630.399553983938,
-      "required_pages_per_second_upper_estimate": 0.002225887633503498
+      "required_token_positions_per_second": 28736.616091201293,
+      "required_pages_per_second_upper_estimate": 0.0022341455020749752
     },
     {
       "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": 28630.399553983938,
-      "required_pages_per_second_upper_estimate": 0.3672400245505309
+      "required_token_positions_per_second": 28736.616091201293,
+      "required_pages_per_second_upper_estimate": 0.36860245624352295
     }
   ],
   "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": 0.09955754881249845,
-    "producer_p95_seconds": 0.2726412000010896,
-    "gpu_update_seconds": 0.9656868374424332,
-    "producer_batches_per_second": 10.04444175180878,
-    "gpu_batches_per_second": 1.0355323912754608,
-    "producer_headroom": 9.699785189178954,
+    "producer_batch_seconds": 0.13055735608600116,
+    "producer_p95_seconds": 0.28683138500127825,
+    "gpu_update_seconds": 0.9621174571234707,
+    "producer_batches_per_second": 7.659468834075322,
+    "gpu_batches_per_second": 1.0393741352431023,
+    "producer_headroom": 7.369308677557024,
     "sustainable": true,
-    "buffer_seconds": 60.83827075887329,
+    "buffer_seconds": 60.61339979877866,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371809.26957304694,
+    "generation_bytes_per_second": 371977.1447808746,
     "upload_bytes_per_second": 4840785.29915486,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.7816139735561,
+    "estimated_full_shards_per_hour": 79.81763608522108,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 787800, "loss": 0.5643776059150696, "lr": 3e-05, "grad_norm": 0.2728983841707396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115311513899, "tokens_per_second": 153320.49250488, "data_wait_s": 9.920000593410805e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 787900, "loss": 0.5382817387580872, "lr": 3e-05, "grad_norm": 0.255097485272613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115326185248, "tokens_per_second": 152130.95550200183, "data_wait_s": 1.150599928223528e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788000, "loss": 0.5319559574127197, "lr": 3e-05, "grad_norm": 0.2482103040428209, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115340762931, "tokens_per_second": 151552.4247322025, "data_wait_s": 1.3317996490513906e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 788100, "loss": 0.5543046593666077, "lr": 3e-05, "grad_norm": 0.2817639190031358, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115355492805, "tokens_per_second": 153133.4143941105, "data_wait_s": 1.506800253991969e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ba2e927d64292f428460333d268e2364334162d4ef7a2d9caa25a7039f7c048
-size 597120
+oid sha256:a495ecb6c741473c8ae9c1f640158bb9f3cf91165f25ebe10356ace56f455217
+size 599238