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": 77520.47418591866,
-      "required_pages_per_second_upper_estimate": 0.005363046276463529
+      "required_token_positions_per_second": 77498.07610866985,
+      "required_pages_per_second_upper_estimate": 0.005361496725508756
     },
     {
       "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": 9690.059273239833,
-      "required_pages_per_second_upper_estimate": 0.004057025466257742
+      "required_token_positions_per_second": 9687.259513583731,
+      "required_pages_per_second_upper_estimate": 0.004055853265355363
     },
     {
       "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": 19380.118546479665,
-      "required_pages_per_second_upper_estimate": 0.004036367605136458
+      "required_token_positions_per_second": 19374.519027167462,
+      "required_pages_per_second_upper_estimate": 0.00403520137293298
     },
     {
       "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": 29070.177819719498,
-      "required_pages_per_second_upper_estimate": 0.0017427127727007781
+      "required_token_positions_per_second": 29061.778540751195,
+      "required_pages_per_second_upper_estimate": 0.0017422092487515846
     },
     {
       "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": 29070.177819719498,
-      "required_pages_per_second_upper_estimate": 0.0022600784592842786
+      "required_token_positions_per_second": 29061.778540751195,
+      "required_pages_per_second_upper_estimate": 0.002259425452289017
     },
     {
       "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": 29070.177819719498,
-      "required_pages_per_second_upper_estimate": 0.37288102794627437
+      "required_token_positions_per_second": 29061.778540751195,
+      "required_pages_per_second_upper_estimate": 0.3727732910141121
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2716447536328168,
-    "producer_p95_seconds": 0.15419631400027356,
-    "gpu_update_seconds": 0.9510777736366381,
-    "producer_batches_per_second": 3.681278532445738,
-    "gpu_batches_per_second": 1.0514387232248081,
-    "producer_headroom": 3.5011821907748435,
+    "producer_batch_seconds": 0.06843660984373656,
+    "producer_p95_seconds": 0.15350703300009627,
+    "gpu_update_seconds": 0.9513526490208176,
+    "producer_batches_per_second": 14.612062203012847,
+    "gpu_batches_per_second": 1.0511349298593458,
+    "producer_headroom": 13.901224084493236,
     "sustainable": true,
-    "buffer_seconds": 60.86897751274484,
+    "buffer_seconds": 60.88656953733233,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404035.1094521738,
-    "upload_bytes_per_second": 4839863.3504238725,
+    "generation_bytes_per_second": 369365.71916065493,
+    "upload_bytes_per_second": 5176279.881902212,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.69652903245841,
+    "estimated_full_shards_per_hour": 79.25728493799912,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 5833727,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.434490552999705,
+      "bytes": 16176474,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 888600, "loss": 0.5856220722198486, "lr": 3e-05, "grad_norm": 0.3560445498797903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130066688750, "tokens_per_second": 153979.4682561426, "data_wait_s": 1.0291999842593214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 888700, "loss": 0.5092302560806274, "lr": 3e-05, "grad_norm": 0.34633805922704586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130081216748, "tokens_per_second": 152221.25253409377, "data_wait_s": 1.2885000160167692e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 888800, "loss": 0.4857497215270996, "lr": 3e-05, "grad_norm": 0.32665121437354316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130095710575, "tokens_per_second": 138544.80322139457, "data_wait_s": 9.969000529963523e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 888900, "loss": 0.5321134924888611, "lr": 3e-05, "grad_norm": 0.31782447520592444, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130110243844, "tokens_per_second": 151452.7059806842, "data_wait_s": 1.4088000170886517e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b293b4b2ea22e67c07a8e10c1abb058465cf5888899eae5f8d680e58be0e1efe
-size 80649
+oid sha256:4e1ae4fa63c488909675184acd00998cfd74a923de685ebff4ffcc682ad98c9b
+size 82767