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": 74360.5659618582,
-      "required_pages_per_second_upper_estimate": 0.0064453853924424705
+      "required_token_positions_per_second": 74934.19872013177,
+      "required_pages_per_second_upper_estimate": 0.00649510642607069
     },
     {
       "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": 9295.070745232275,
-      "required_pages_per_second_upper_estimate": 0.003707942693965324
+      "required_token_positions_per_second": 9366.774840016471,
+      "required_pages_per_second_upper_estimate": 0.003736546529446494
     },
     {
       "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": 18590.14149046455,
-      "required_pages_per_second_upper_estimate": 0.004300502058503166
+      "required_token_positions_per_second": 18733.549680032942,
+      "required_pages_per_second_upper_estimate": 0.004333677019261879
     },
     {
       "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": 27885.212235696825,
-      "required_pages_per_second_upper_estimate": 0.0016980631227521813
+      "required_token_positions_per_second": 28100.324520049413,
+      "required_pages_per_second_upper_estimate": 0.0017111623322623212
     },
     {
       "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": 27885.212235696825,
-      "required_pages_per_second_upper_estimate": 0.002146680829111867
+      "required_token_positions_per_second": 28100.324520049413,
+      "required_pages_per_second_upper_estimate": 0.002163240768230244
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27885.212235696825,
-      "required_pages_per_second_upper_estimate": 0.34624101016547454
+      "required_token_positions_per_second": 28100.324520049413,
+      "required_pages_per_second_upper_estimate": 0.34891198480263
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09906495359376066,
+    "producer_batch_seconds": 0.04651564814843212,
     "producer_p95_seconds": 0.1561405239999658,
-    "gpu_update_seconds": 0.9914932605249043,
-    "producer_batches_per_second": 10.094387204790275,
-    "gpu_batches_per_second": 1.0085797249600992,
-    "producer_headroom": 10.008516882678384,
+    "gpu_update_seconds": 0.9839032278888209,
+    "producer_batches_per_second": 21.498141804000777,
+    "gpu_batches_per_second": 1.0163601171892873,
+    "producer_headroom": 21.152091114567966,
     "sustainable": true,
-    "buffer_seconds": 60.48108889201916,
+    "buffer_seconds": 60.01809690121807,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299661.2107092799,
-    "upload_bytes_per_second": 4560733.846997692,
+    "generation_bytes_per_second": 311593.14407419745,
+    "upload_bytes_per_second": 4633958.315790001,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.3003200622444,
+    "estimated_full_shards_per_hour": 66.86063520116275,
     "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": [
@@ -274,6 +274,12 @@
       "bytes": 1038671,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.359243317000164,
+      "bytes": 7794752,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -19,3 +19,4 @@
 {"step": 316900, "loss": 0.5719795078039169, "lr": 0.0003, "grad_norm": 0.1662062516579049, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46385720159, "tokens_per_second": 150902.1330163642, "data_wait_s": 1.2445999800547725e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 317000, "loss": 0.6288293264806271, "lr": 0.0003, "grad_norm": 0.17431565773266877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46400491994, "tokens_per_second": 148827.53083358146, "data_wait_s": 1.2710000191873405e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 317100, "loss": 0.5828649736940861, "lr": 0.0003, "grad_norm": 0.1620742490346258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46415209341, "tokens_per_second": 149288.2822589277, "data_wait_s": 1.3791999663226306e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 317200, "loss": 0.604051448404789, "lr": 0.0003, "grad_norm": 0.1369729481549036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46429715538, "tokens_per_second": 121843.6044977407, "data_wait_s": 1.4817000192124397e-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 sha256:71c5f2d53895d2ac9a5a56f388fa9528329b1b4cf00a18d8dbe173f83e4e51e1
-size 44643
+oid sha256:f2de3da79790af4b06a57e5ffab88f6ca5d66316d27c3f2c7a493943bbbb814b
+size 46761