Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75764.0573672814,
-      "required_pages_per_second_upper_estimate": 0.006567036470348108
+      "required_token_positions_per_second": 75709.46314947892,
+      "required_pages_per_second_upper_estimate": 0.006562304381916768
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9470.507170910176,
-      "required_pages_per_second_upper_estimate": 0.00377792690717655
+      "required_token_positions_per_second": 9463.682893684865,
+      "required_pages_per_second_upper_estimate": 0.003775204600959337
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18941.01434182035,
-      "required_pages_per_second_upper_estimate": 0.004381670317513055
+      "required_token_positions_per_second": 18927.36578736973,
+      "required_pages_per_second_upper_estimate": 0.004378512964647257
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28411.521512730527,
-      "required_pages_per_second_upper_estimate": 0.0017301125963921642
+      "required_token_positions_per_second": 28391.048681054595,
+      "required_pages_per_second_upper_estimate": 0.0017288659083557463
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28411.521512730527,
-      "required_pages_per_second_upper_estimate": 0.0021871975741752474
+      "required_token_positions_per_second": 28391.048681054595,
+      "required_pages_per_second_upper_estimate": 0.002185621518920409
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28411.521512730527,
-      "required_pages_per_second_upper_estimate": 0.3527760099423933
+      "required_token_positions_per_second": 28391.048681054595,
+      "required_pages_per_second_upper_estimate": 0.3525218058911382
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.39021835360932755,
-    "producer_p95_seconds": 0.21085902099912346,
-    "gpu_update_seconds": 0.9731263419881118,
-    "producer_batches_per_second": 2.5626677749790407,
-    "gpu_batches_per_second": 1.0276157954546632,
-    "producer_headroom": 2.4937995175961674,
+    "producer_batch_seconds": 0.08561632525784546,
+    "producer_p95_seconds": 0.2871364149996225,
+    "gpu_update_seconds": 0.9738280649861858,
+    "producer_batches_per_second": 11.680015429164484,
+    "gpu_batches_per_second": 1.026875313984903,
+    "producer_headroom": 11.374326824392044,
     "sustainable": true,
-    "buffer_seconds": 60.33383320326293,
+    "buffer_seconds": 60.37734002914352,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362373.10742970754,
-    "upload_bytes_per_second": 4051490.818543762,
+    "generation_bytes_per_second": 366747.46536990005,
+    "upload_bytes_per_second": 4137697.6116614225,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.75683324020785,
+    "estimated_full_shards_per_hour": 78.69546862433197,
     "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": 1.1734728310002538,
-      "bytes": 6054797,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2160326559996975,
       "bytes": 2369678,
@@ -562,6 +556,12 @@
       "bytes": 1078743,
       "pages": 30,
       "shards": 1
+    },
+    {
+      "seconds": 1.3106396379989746,
+      "bytes": 15458880,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 329400, "loss": 0.6524562239646912, "lr": 0.0003, "grad_norm": 0.15385772274837856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48215914009, "tokens_per_second": 149490.48071541806, "data_wait_s": 1.3350001609069295e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329500, "loss": 0.603059608489275, "lr": 0.0003, "grad_norm": 0.1480292167152202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48230511147, "tokens_per_second": 149630.44984669285, "data_wait_s": 1.3916000170866027e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329600, "loss": 0.6107210144400597, "lr": 0.0003, "grad_norm": 0.144386048113135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48245021257, "tokens_per_second": 146190.4000532596, "data_wait_s": 1.471900031901896e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 329700, "loss": 0.5373153388500214, "lr": 0.0003, "grad_norm": 0.15336776926794268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48259548074, "tokens_per_second": 147798.01532501896, "data_wait_s": 1.5064999388414435e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0ea9d35aa65442f2b997efe0ba3bf188d8c994d21811e448fa11313ee9d44964
-size 310959
+oid sha256:9b23c8bd21890ccf828639baba1747f5a794e0f7d29cff5574f2f1f49df2e3ab
+size 313077