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": 76565.79414645793,
-      "required_pages_per_second_upper_estimate": 0.005296999296170754
+      "required_token_positions_per_second": 77354.96716780533,
+      "required_pages_per_second_upper_estimate": 0.0053515961169212575
     },
     {
       "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": 9570.724268307242,
-      "required_pages_per_second_upper_estimate": 0.0040070623916907445
+      "required_token_positions_per_second": 9669.370895975666,
+      "required_pages_per_second_upper_estimate": 0.0040483636747196804
     },
     {
       "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": 19141.448536614484,
-      "required_pages_per_second_upper_estimate": 0.003986658936232964
+      "required_token_positions_per_second": 19338.741791951332,
+      "required_pages_per_second_upper_estimate": 0.004027749918346602
     },
     {
       "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": 28712.172804921724,
-      "required_pages_per_second_upper_estimate": 0.0017212509187056575
+      "required_token_positions_per_second": 29008.112687926994,
+      "required_pages_per_second_upper_estimate": 0.0017389920628177865
     },
     {
       "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": 28712.172804921724,
-      "required_pages_per_second_upper_estimate": 0.0022322451440813925
+      "required_token_positions_per_second": 29008.112687926994,
+      "required_pages_per_second_upper_estimate": 0.002255253168283075
     },
     {
       "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": 28712.172804921724,
-      "required_pages_per_second_upper_estimate": 0.36828892401228464
+      "required_token_positions_per_second": 29008.112687926994,
+      "required_pages_per_second_upper_estimate": 0.3720849230759866
     }
   ],
   "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": 0.067691734976421,
-    "producer_p95_seconds": 0.24887008000223432,
-    "gpu_update_seconds": 0.9629365282748887,
-    "producer_batches_per_second": 14.772852259560626,
-    "gpu_batches_per_second": 1.03849004647431,
-    "producer_headroom": 14.225319067539154,
+    "producer_batch_seconds": 0.135684689000243,
+    "producer_p95_seconds": 0.2593115769996075,
+    "gpu_update_seconds": 0.9531126791129343,
+    "producer_batches_per_second": 7.370028316151494,
+    "gpu_batches_per_second": 1.0491938906223595,
+    "producer_headroom": 7.024467433545338,
     "sustainable": true,
-    "buffer_seconds": 60.66500128131799,
+    "buffer_seconds": 60.04609878411486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346296.08435846475,
-    "upload_bytes_per_second": 5025204.241395261,
+    "generation_bytes_per_second": 356398.28278081416,
+    "upload_bytes_per_second": 5012812.619410316,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.30707834306199,
+    "estimated_full_shards_per_hour": 76.47477495735473,
     "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": 1.3964128229999915,
-      "bytes": 12688722,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.4358708510008,
       "bytes": 6164418,
@@ -562,6 +556,12 @@
       "bytes": 7721145,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3206068420004158,
+      "bytes": 11244076,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -198,3 +198,4 @@
 {"step": 904800, "loss": 0.5306363701820374, "lr": 3e-05, "grad_norm": 0.3199387431222164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132436887047, "tokens_per_second": 153729.73392596567, "data_wait_s": 1.7258000298170373e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 904900, "loss": 0.5055314302444458, "lr": 3e-05, "grad_norm": 0.3194904783678648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132451574646, "tokens_per_second": 152710.08105990206, "data_wait_s": 1.3526998372981325e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 905000, "loss": 0.5308002829551697, "lr": 3e-05, "grad_norm": 0.3205761780646895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132466342829, "tokens_per_second": 154075.54590951814, "data_wait_s": 1.4929999451851472e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 905100, "loss": 0.5184240341186523, "lr": 3e-05, "grad_norm": 0.3432106213918256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132481075187, "tokens_per_second": 142521.79365336325, "data_wait_s": 1.6055997548392043e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4904c9794151f0b6582da0f467cf1c60564851973e8924130a74996f50812dbd
-size 426897
+oid sha256:8450c2f6434517a91b699cd32b810f77f69df166a446ffe7dfa6cc4ec43ec744
+size 429015