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": 77224.56101832613,
-      "required_pages_per_second_upper_estimate": 0.005342574316916332
+      "required_token_positions_per_second": 76150.08499308299,
+      "required_pages_per_second_upper_estimate": 0.0052682395723621465
     },
     {
       "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": 9653.070127290766,
-      "required_pages_per_second_upper_estimate": 0.004041538883270071
+      "required_token_positions_per_second": 9518.760624135373,
+      "required_pages_per_second_upper_estimate": 0.0039853062990002665
     },
     {
       "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": 19306.140254581533,
-      "required_pages_per_second_upper_estimate": 0.004020959877872829
+      "required_token_positions_per_second": 19037.521248270747,
+      "required_pages_per_second_upper_estimate": 0.003965013622817864
     },
     {
       "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": 28959.210381872297,
-      "required_pages_per_second_upper_estimate": 0.0017360604442393053
+      "required_token_positions_per_second": 28556.28187240612,
+      "required_pages_per_second_upper_estimate": 0.0017119054953330186
     },
     {
       "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": 28959.210381872297,
-      "required_pages_per_second_upper_estimate": 0.0022514512290858343
+      "required_token_positions_per_second": 28556.28187240612,
+      "required_pages_per_second_upper_estimate": 0.00222012530977005
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28959.210381872297,
-      "required_pages_per_second_upper_estimate": 0.3714576568011223
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28556.28187240612,
+      "required_pages_per_second_upper_estimate": 0.36628932251261687
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09809735991410662,
-    "producer_p95_seconds": 0.22224198899857583,
-    "gpu_update_seconds": 0.9547221638787127,
-    "producer_batches_per_second": 10.193954260090111,
-    "gpu_batches_per_second": 1.0474251440202653,
-    "producer_headroom": 9.732394069673852,
+    "producer_batch_seconds": 0.12527987708580213,
+    "producer_p95_seconds": 0.24405083400051808,
+    "gpu_update_seconds": 0.9681932726233595,
+    "producer_batches_per_second": 7.982127882478017,
+    "gpu_batches_per_second": 1.0328516302230224,
+    "producer_headroom": 7.728242517034558,
     "sustainable": true,
-    "buffer_seconds": 60.1474963243589,
+    "buffer_seconds": 60.02798290264829,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 218006.7285878414,
-    "upload_bytes_per_second": 5147180.443778019,
+    "generation_bytes_per_second": 224947.25220927532,
+    "upload_bytes_per_second": 5135121.386962364,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 46.779169018043824,
+    "estimated_full_shards_per_hour": 48.26844382008262,
     "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": 0.9391229799985013,
-      "bytes": 2632765,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.0898778940008924,
       "bytes": 10697456,
@@ -562,6 +556,12 @@
       "bytes": 12018438,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.0833680409996305,
+      "bytes": 2366981,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -186,3 +186,4 @@
 {"step": 575700, "loss": 0.5894057154655457, "lr": 0.0003, "grad_norm": 0.15861235980839236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84265001282, "tokens_per_second": 67196.43554143806, "data_wait_s": 7.2690017987042665e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 575800, "loss": 0.546024739742279, "lr": 0.0003, "grad_norm": 0.18850383590792041, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84279627659, "tokens_per_second": 152615.38952956314, "data_wait_s": 1.4261000615078956e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 575900, "loss": 0.5540058612823486, "lr": 0.0003, "grad_norm": 0.1559572572497929, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84294224818, "tokens_per_second": 151778.02242146767, "data_wait_s": 1.5068999346112832e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 576000, "loss": 0.6009294986724854, "lr": 0.0003, "grad_norm": 0.15569686658394694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 84309105577, "tokens_per_second": 154206.54879942816, "data_wait_s": 1.1015999916708097e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd153f121ff5424f0acdb692373736381d6f9f2be845704215bea4dc17e2c26a
-size 401481
+oid sha256:3a24957bd06223620e81f426932e3aea9a7047084e2cafa3e715ded0bbb4d956
+size 403599