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": 77318.76817415735,
-      "required_pages_per_second_upper_estimate": 0.005349091786547452
+      "required_token_positions_per_second": 77369.052332473,
+      "required_pages_per_second_upper_estimate": 0.005352570561295051
     },
     {
       "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": 9664.846021769668,
-      "required_pages_per_second_upper_estimate": 0.004046469204379747
+      "required_token_positions_per_second": 9671.131541559125,
+      "required_pages_per_second_upper_estimate": 0.004049100820259014
     },
     {
       "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": 19329.692043539337,
-      "required_pages_per_second_upper_estimate": 0.004025865094410297
+      "required_token_positions_per_second": 19342.26308311825,
+      "required_pages_per_second_upper_estimate": 0.004028483310433977
     },
     {
       "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": 28994.538065309007,
-      "required_pages_per_second_upper_estimate": 0.0017381782849190867
+      "required_token_positions_per_second": 29013.394624677378,
+      "required_pages_per_second_upper_estimate": 0.0017393087068609215
     },
     {
       "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": 28994.538065309007,
-      "required_pages_per_second_upper_estimate": 0.002254197801082981
+      "required_token_positions_per_second": 29013.394624677378,
+      "required_pages_per_second_upper_estimate": 0.002255663815632633
     },
     {
       "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": 28994.538065309007,
-      "required_pages_per_second_upper_estimate": 0.37191080239233726
+      "required_token_positions_per_second": 29013.394624677378,
+      "required_pages_per_second_upper_estimate": 0.37215267408931874
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09672759722661795,
-    "producer_p95_seconds": 0.2603394619982282,
-    "gpu_update_seconds": 0.9535589060851397,
-    "producer_batches_per_second": 10.338311181835243,
-    "gpu_batches_per_second": 1.0487029103482715,
-    "producer_headroom": 9.858188701318582,
+    "producer_batch_seconds": 0.08060388910178062,
+    "producer_p95_seconds": 0.22325702699890826,
+    "gpu_update_seconds": 0.9529391633643572,
+    "producer_batches_per_second": 12.40634926110419,
+    "gpu_batches_per_second": 1.0493849328948703,
+    "producer_headroom": 11.822496085282637,
     "sustainable": true,
-    "buffer_seconds": 60.0742110833638,
+    "buffer_seconds": 60.03516729195451,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352998.4956773428,
+    "generation_bytes_per_second": 384295.560956119,
     "upload_bytes_per_second": 5552634.537313602,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.74525978794301,
+    "estimated_full_shards_per_hour": 82.4608814383762,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -173,3 +173,4 @@
 {"step": 502300, "loss": 0.5867668390274048, "lr": 0.0003, "grad_norm": 0.153646522754659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73519812847, "tokens_per_second": 151311.08754722233, "data_wait_s": 1.3714998203795403e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502400, "loss": 0.6325347423553467, "lr": 0.0003, "grad_norm": 0.20552976388713154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73534579820, "tokens_per_second": 153798.08455313052, "data_wait_s": 9.646999387769029e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 502500, "loss": 0.5735471248626709, "lr": 0.0003, "grad_norm": 0.15704754788065672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73549106769, "tokens_per_second": 150982.9314653788, "data_wait_s": 1.1591000657062978e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 502600, "loss": 0.5724793076515198, "lr": 0.0003, "grad_norm": 0.1434232357324958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73563901469, "tokens_per_second": 154233.70855758447, "data_wait_s": 1.748100112308748e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3c215fe13c860eb970890ac05a7acea9b67f67c3bbc87ebc76116b968500b1e3
-size 373164
+oid sha256:3b44465dc2bf3f0c0a39d240d3bf42c7e22a6472ec626f052f78a2262d16d3d7
+size 375282