Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75075.94406547502,
-      "required_pages_per_second_upper_estimate": 0.006507392553354749
+      "required_token_positions_per_second": 74989.20060169409,
+      "required_pages_per_second_upper_estimate": 0.006499873849763517
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9384.493008184378,
-      "required_pages_per_second_upper_estimate": 0.0037436145716388935
+      "required_token_positions_per_second": 9373.65007521176,
+      "required_pages_per_second_upper_estimate": 0.0037392891635598217
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18768.986016368755,
-      "required_pages_per_second_upper_estimate": 0.00434187459201495
+      "required_token_positions_per_second": 18747.30015042352,
+      "required_pages_per_second_upper_estimate": 0.004336857948586726
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28153.479024553133,
-      "required_pages_per_second_upper_estimate": 0.0017143991627064656
+      "required_token_positions_per_second": 28120.950225635286,
+      "required_pages_per_second_upper_estimate": 0.001712418329517787
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28153.479024553133,
-      "required_pages_per_second_upper_estimate": 0.0021673327491280262
+      "required_token_positions_per_second": 28120.950225635286,
+      "required_pages_per_second_upper_estimate": 0.002164828592141852
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28153.479024553133,
-      "required_pages_per_second_upper_estimate": 0.34957198585188337
+      "required_token_positions_per_second": 28120.950225635286,
+      "required_pages_per_second_upper_estimate": 0.3491680870362105
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.33574176570306236,
-    "producer_p95_seconds": 0.35833659800118767,
-    "gpu_update_seconds": 0.9820455928692757,
-    "producer_batches_per_second": 2.9784795999566605,
-    "gpu_batches_per_second": 1.0182826614783396,
-    "producer_headroom": 2.925002764588482,
+    "producer_batch_seconds": 0.11951990499193244,
+    "producer_p95_seconds": 0.36735511800361564,
+    "gpu_update_seconds": 0.9831815702584566,
+    "producer_batches_per_second": 8.366807186363642,
+    "gpu_batches_per_second": 1.0171061279526652,
+    "producer_headroom": 8.226090627538746,
     "sustainable": true,
-    "buffer_seconds": 60.88682675789509,
+    "buffer_seconds": 60.957257356024314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352990.1561887343,
-    "upload_bytes_per_second": 3821498.7403385686,
+    "generation_bytes_per_second": 353171.3736934166,
+    "upload_bytes_per_second": 5421713.68600837,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.74347032782099,
+    "estimated_full_shards_per_hour": 75.7823553857982,
     "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": 34.0496005649984,
-      "bytes": 11526687,
-      "pages": 38,
-      "shards": 4
-    },
     {
       "seconds": 1.0691167909972137,
       "bytes": 6134154,
@@ -562,6 +556,12 @@
       "bytes": 5999613,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.4630355209956178,
+      "bytes": 11583159,
+      "pages": 30,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -389,3 +389,4 @@
 {"step": 353900, "loss": 0.6297894176095724, "lr": 0.0003, "grad_norm": 0.160765271161839, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51801038414, "tokens_per_second": 150042.8605037509, "data_wait_s": 1.3095996109768748e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354000, "loss": 0.6086098346859217, "lr": 0.0003, "grad_norm": 0.1419565087488861, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51815734487, "tokens_per_second": 149807.99207376977, "data_wait_s": 1.6703001165296882e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 354100, "loss": 0.5691184941679239, "lr": 0.0003, "grad_norm": 0.1687424969750012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51830503624, "tokens_per_second": 149911.55515781548, "data_wait_s": 1.4862001989968121e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 354200, "loss": 0.5616220980882645, "lr": 0.0003, "grad_norm": 0.13573110957447848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51845169975, "tokens_per_second": 148393.61423746162, "data_wait_s": 1.0702999134082347e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3d81f1a3f00d262ccc3203154a8d1246b746ac380fa2e849dde1bb5845a703c
-size 833784
+oid sha256:cd0a3e861865dffba3975fcd875a64ee5e06f4df4da724e6d8e7867bd3658de7
+size 835902