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": 76706.37329300144,
-      "required_pages_per_second_upper_estimate": 0.005306724887717194
+      "required_token_positions_per_second": 76458.25367083235,
+      "required_pages_per_second_upper_estimate": 0.005289559396538694
     },
     {
       "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": 9588.29666162518,
-      "required_pages_per_second_upper_estimate": 0.0040144195858196645
+      "required_token_positions_per_second": 9557.281708854043,
+      "required_pages_per_second_upper_estimate": 0.004001434272760253
     },
     {
       "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": 19176.59332325036,
-      "required_pages_per_second_upper_estimate": 0.003993978668458867
+      "required_token_positions_per_second": 19114.563417708086,
+      "required_pages_per_second_upper_estimate": 0.003981059474973026
     },
     {
       "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": 28764.889984875535,
-      "required_pages_per_second_upper_estimate": 0.0017244112331494155
+      "required_token_positions_per_second": 28671.845126562133,
+      "required_pages_per_second_upper_estimate": 0.0017188333359647995
     },
     {
       "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": 28764.889984875535,
-      "required_pages_per_second_upper_estimate": 0.0022363436729444324
+      "required_token_positions_per_second": 28671.845126562133,
+      "required_pages_per_second_upper_estimate": 0.002229109844471645
     },
     {
       "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": 28764.889984875535,
-      "required_pages_per_second_upper_estimate": 0.36896512339343435
+      "required_token_positions_per_second": 28671.845126562133,
+      "required_pages_per_second_upper_estimate": 0.36777164385477523
     }
   ],
   "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": 0.16092992855479338,
-    "producer_p95_seconds": 0.27221962900148355,
-    "gpu_update_seconds": 0.9611717623302994,
-    "producer_batches_per_second": 6.213884570634854,
-    "gpu_batches_per_second": 1.0403967731798154,
-    "producer_headroom": 5.972610383674159,
+    "producer_batch_seconds": 0.0676338518044588,
+    "producer_p95_seconds": 0.25433909200000926,
+    "gpu_update_seconds": 0.96429092295795,
+    "producer_batches_per_second": 14.785495329929951,
+    "gpu_batches_per_second": 1.0370314354225307,
+    "producer_headroom": 14.257518938088612,
     "sustainable": true,
-    "buffer_seconds": 60.55382102680886,
+    "buffer_seconds": 60.75032814635085,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365441.9612129554,
-    "upload_bytes_per_second": 4983297.93687411,
+    "generation_bytes_per_second": 427516.4032326214,
+    "upload_bytes_per_second": 5051899.891187658,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.41533782283304,
+    "estimated_full_shards_per_hour": 91.735068061199,
     "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": 1.5897931969993806,
-      "bytes": 7707081,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0018968719996337,
       "bytes": 927002,
@@ -562,6 +556,12 @@
       "bytes": 2347277,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2341140790013014,
+      "bytes": 11781395,
+      "pages": 31,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -201,3 +201,4 @@
 {"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}
 {"step": 905200, "loss": 0.5125197768211365, "lr": 3e-05, "grad_norm": 0.3015187881744235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132495800874, "tokens_per_second": 153105.42242740648, "data_wait_s": 1.251199864782393e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 905300, "loss": 0.5703348517417908, "lr": 3e-05, "grad_norm": 0.31669086345190606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132510504768, "tokens_per_second": 153165.9361834013, "data_wait_s": 1.2901997251901776e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 905400, "loss": 0.542047381401062, "lr": 3e-05, "grad_norm": 0.33252034845310646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132525219983, "tokens_per_second": 153165.92505290068, "data_wait_s": 1.596699803485535e-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:b628d507a0e04002ff8df7704e262487c3ba2af378c4b88bd8fdfe3b4bc9c431
-size 433251
+oid sha256:f1ca08a182fe3748e20e8ab50ca448b0767ad077fdf2b36b7caf8b82713374d9
+size 435369