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": 76932.58955511708,
-      "required_pages_per_second_upper_estimate": 0.005322375053624398
+      "required_token_positions_per_second": 76770.44018312877,
+      "required_pages_per_second_upper_estimate": 0.0053111571838318685
     },
     {
       "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": 9616.573694389635,
-      "required_pages_per_second_upper_estimate": 0.004026258588946551
+      "required_token_positions_per_second": 9596.305022891096,
+      "required_pages_per_second_upper_estimate": 0.004017772519448104
     },
     {
       "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": 19233.14738877927,
-      "required_pages_per_second_upper_estimate": 0.0040057573888775365
+      "required_token_positions_per_second": 19192.610045782192,
+      "required_pages_per_second_upper_estimate": 0.003997314529372869
     },
     {
       "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": 28849.721083168904,
-      "required_pages_per_second_upper_estimate": 0.0017294967279625159
+      "required_token_positions_per_second": 28788.915068673286,
+      "required_pages_per_second_upper_estimate": 0.0017258515002389117
     },
     {
       "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": 28849.721083168904,
-      "required_pages_per_second_upper_estimate": 0.0022429389177041742
+      "required_token_positions_per_second": 28788.915068673286,
+      "required_pages_per_second_upper_estimate": 0.002238211517534532
     },
     {
       "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": 28849.721083168904,
-      "required_pages_per_second_upper_estimate": 0.37005324563780484
+      "required_token_positions_per_second": 28788.915068673286,
+      "required_pages_per_second_upper_estimate": 0.3692732913722667
     }
   ],
   "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": 0.09394030317977808,
-    "producer_p95_seconds": 0.26087649499822874,
-    "gpu_update_seconds": 0.9583454869561981,
-    "producer_batches_per_second": 10.645058256691506,
-    "gpu_batches_per_second": 1.043465027603042,
-    "producer_headroom": 10.201643538686119,
+    "producer_batch_seconds": 0.10356374351542286,
+    "producer_p95_seconds": 0.24648861500099883,
+    "gpu_update_seconds": 0.9603696399828983,
+    "producer_batches_per_second": 9.655888885969814,
+    "gpu_batches_per_second": 1.0412657359907873,
+    "producer_headroom": 9.2732225331337,
     "sustainable": true,
-    "buffer_seconds": 60.375765678240484,
+    "buffer_seconds": 60.50328731892259,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 161,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 571623.1440471476,
+    "generation_bytes_per_second": 562024.8503533086,
     "upload_bytes_per_second": 4973413.749067708,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.32851000337992,
+    "estimated_full_shards_per_hour": 60.29872480845186,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 906900, "loss": 0.5580467581748962, "lr": 3e-05, "grad_norm": 0.3092025897628717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132744019053, "tokens_per_second": 151957.30139648938, "data_wait_s": 1.499800055171363e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907000, "loss": 0.5452108979225159, "lr": 3e-05, "grad_norm": 0.32432525863845285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132758583743, "tokens_per_second": 151527.62240822273, "data_wait_s": 1.5677996998419985e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 907100, "loss": 0.4869060814380646, "lr": 3e-05, "grad_norm": 0.32425706873955157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132773278589, "tokens_per_second": 152844.0312618162, "data_wait_s": 1.7708996892906725e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 907200, "loss": 0.5460553169250488, "lr": 3e-05, "grad_norm": 0.315383454066576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132787906599, "tokens_per_second": 152128.29741458583, "data_wait_s": 1.3246997696114704e-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 sha256:84c3b7c660b62b188505274ee16f6ac8591d4c61034de78b3bf001ca63397da5
-size 471375
+oid sha256:a1df27014bc1c36163ad59b1c897dd6b2b7bae6f9d6045a9cde94ac022ced66d
+size 473493