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": 74991.7627555206,
-      "required_pages_per_second_upper_estimate": 0.0051881041522897
+      "required_token_positions_per_second": 75334.75334039006,
+      "required_pages_per_second_upper_estimate": 0.005211833036798772
     },
     {
       "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": 9373.970344440075,
-      "required_pages_per_second_upper_estimate": 0.003924685632456707
+      "required_token_positions_per_second": 9416.844167548757,
+      "required_pages_per_second_upper_estimate": 0.003942636006885077
     },
     {
       "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": 18747.94068888015,
-      "required_pages_per_second_upper_estimate": 0.0039047016290497034
+      "required_token_positions_per_second": 18833.688335097515,
+      "required_pages_per_second_upper_estimate": 0.003922560602439283
     },
     {
       "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": 28121.911033320222,
-      "required_pages_per_second_upper_estimate": 0.001685865626775687
+      "required_token_positions_per_second": 28250.532502646274,
+      "required_pages_per_second_upper_estimate": 0.001693576287468172
     },
     {
       "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": 28121.911033320222,
-      "required_pages_per_second_upper_estimate": 0.002186354887626525
+      "required_token_positions_per_second": 28250.532502646274,
+      "required_pages_per_second_upper_estimate": 0.0021963546411205718
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28121.911033320222,
-      "required_pages_per_second_upper_estimate": 0.3607176797798928
+      "required_token_positions_per_second": 28250.532502646274,
+      "required_pages_per_second_upper_estimate": 0.36236749788543343
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.3476239325941606,
-    "producer_p95_seconds": 0.35883774599642493,
-    "gpu_update_seconds": 0.9831479790701738,
-    "producer_batches_per_second": 2.8766719038515305,
-    "gpu_batches_per_second": 1.0171408793880288,
-    "producer_headroom": 2.8281941687195817,
+    "producer_batch_seconds": 0.1048541820704827,
+    "producer_p95_seconds": 0.3682760690062423,
+    "gpu_update_seconds": 0.978671817864324,
+    "producer_batches_per_second": 9.537054033074263,
+    "gpu_batches_per_second": 1.0217929869302038,
+    "producer_headroom": 9.333646007619071,
     "sustainable": true,
-    "buffer_seconds": 60.955174702350774,
+    "buffer_seconds": 60.67765270758809,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 650252.5945651984,
-    "upload_bytes_per_second": 4960399.819160943,
+    "generation_bytes_per_second": 653417.8932605253,
+    "upload_bytes_per_second": 5038073.572900272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.76453484400255,
+    "estimated_full_shards_per_hour": 70.1041345518199,
     "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": 1.1499276460017427,
-      "bytes": 7228890,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.1513834180004778,
       "bytes": 6077960,
@@ -562,6 +556,12 @@
       "bytes": 2453796,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.5481451480009127,
+      "bytes": 15819770,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -424,3 +424,4 @@
 {"step": 967400, "loss": 0.5333903431892395, "lr": 3e-05, "grad_norm": 0.3574164218272315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141601135050, "tokens_per_second": 153021.78144090463, "data_wait_s": 1.2317999789956957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 967500, "loss": 0.5437011122703552, "lr": 3e-05, "grad_norm": 0.3618753728928093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141615881872, "tokens_per_second": 151832.95373416174, "data_wait_s": 1.6120000509545207e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 967600, "loss": 0.5181931257247925, "lr": 3e-05, "grad_norm": 0.3292315335811378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141630607289, "tokens_per_second": 152958.61733542618, "data_wait_s": 8.55499820318073e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 967700, "loss": 0.507817804813385, "lr": 3e-05, "grad_norm": 0.34821866907319726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141645206040, "tokens_per_second": 151648.98401438427, "data_wait_s": 1.2882002920378e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39b800147c62dbdceff1e7cc1f0d6fed7577a7f2e5c2bbcf68d24adbeb163896
-size 908697
+oid sha256:0694d7410699f572bc62d4080786bd54dcae850cd1cacdcc5c46c2fd7edd193d
+size 910815