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": 75732.11169568915,
-      "required_pages_per_second_upper_estimate": 0.005239323209816774
+      "required_token_positions_per_second": 77321.13215342937,
+      "required_pages_per_second_upper_estimate": 0.005349255332118679
     },
     {
       "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": 9466.513961961144,
-      "required_pages_per_second_upper_estimate": 0.003963431712582289
+      "required_token_positions_per_second": 9665.141519178671,
+      "required_pages_per_second_upper_estimate": 0.004046592922974209
     },
     {
       "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": 18933.027923922287,
-      "required_pages_per_second_upper_estimate": 0.003943250419030354
+      "required_token_positions_per_second": 19330.283038357342,
+      "required_pages_per_second_upper_estimate": 0.004025988183045307
     },
     {
       "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": 28399.54188588343,
-      "required_pages_per_second_upper_estimate": 0.0017025091724690853
+      "required_token_positions_per_second": 28995.424557536015,
+      "required_pages_per_second_upper_estimate": 0.0017382314287745005
     },
     {
       "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": 28399.54188588343,
-      "required_pages_per_second_upper_estimate": 0.002207939465244251
+      "required_token_positions_per_second": 28995.424557536015,
+      "required_pages_per_second_upper_estimate": 0.002254266721954364
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28399.54188588343,
-      "required_pages_per_second_upper_estimate": 0.36427883025978924
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28995.424557536015,
+      "required_pages_per_second_upper_estimate": 0.37192217336278416
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.5375210356484388,
-    "producer_p95_seconds": 0.26334111699907226,
-    "gpu_update_seconds": 0.9735368306677862,
-    "producer_batches_per_second": 1.8603923077980558,
-    "gpu_batches_per_second": 1.0271825045530756,
-    "producer_headroom": 1.8111604311324476,
+    "producer_batch_seconds": 0.08750120275789186,
+    "producer_p95_seconds": 0.2603394619982282,
+    "gpu_update_seconds": 0.9535297524317224,
+    "producer_batches_per_second": 11.42841433582247,
+    "gpu_batches_per_second": 1.0487349738692135,
+    "producer_headroom": 10.897333092323947,
     "sustainable": true,
-    "buffer_seconds": 60.35928350140274,
+    "buffer_seconds": 60.072374403198516,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352339.9428881008,
-    "upload_bytes_per_second": 5536506.73393112,
+    "generation_bytes_per_second": 352805.1381971763,
+    "upload_bytes_per_second": 5615311.664059876,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.60394968969601,
+    "estimated_full_shards_per_hour": 75.70376977383104,
     "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": 1.1789036200007104,
-      "bytes": 2313669,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.5398991070014745,
       "bytes": 10300581,
@@ -562,6 +556,12 @@
       "bytes": 13434552,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.516792244001408,
+      "bytes": 10349687,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -171,3 +171,4 @@
 {"step": 502100, "loss": 0.5456511974334717, "lr": 0.0003, "grad_norm": 0.1837708465292662, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73490554006, "tokens_per_second": 152031.23970780458, "data_wait_s": 9.715000487631187e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 502200, "loss": 0.626147985458374, "lr": 0.0003, "grad_norm": 0.2118523773437156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73505191055, "tokens_per_second": 136789.732625182, "data_wait_s": 11.220147740998073, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:45740a1277506c44cb64310ed68f78a6fefda8d729ac9ecf267c295f31f5f23f
-size 368928
+oid sha256:de9514b8dbbd38596957255beb48c2746f297100bc0b3e396f8b9303c861031a
+size 371046