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": 77355.08767681957,
-      "required_pages_per_second_upper_estimate": 0.0053516044540144825
+      "required_token_positions_per_second": 77062.68958313447,
+      "required_pages_per_second_upper_estimate": 0.0053313756754362965
     },
     {
       "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": 9669.385959602447,
-      "required_pages_per_second_upper_estimate": 0.004048369981545649
+      "required_token_positions_per_second": 9632.83619789181,
+      "required_pages_per_second_upper_estimate": 0.004033067359563221
     },
     {
       "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": 19338.771919204893,
-      "required_pages_per_second_upper_estimate": 0.004027756193059009
+      "required_token_positions_per_second": 19265.67239578362,
+      "required_pages_per_second_upper_estimate": 0.004012531490094427
     },
     {
       "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": 29008.157878807342,
-      "required_pages_per_second_upper_estimate": 0.0017389947719420608
+      "required_token_positions_per_second": 28898.508593675426,
+      "required_pages_per_second_upper_estimate": 0.0017324214647231667
     },
     {
       "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": 29008.157878807342,
-      "required_pages_per_second_upper_estimate": 0.0022552566816752486
+      "required_token_positions_per_second": 28898.508593675426,
+      "required_pages_per_second_upper_estimate": 0.0022467319320524765
     },
     {
       "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": 29008.157878807342,
-      "required_pages_per_second_upper_estimate": 0.37208550273607754
+      "required_token_positions_per_second": 28898.508593675426,
+      "required_pages_per_second_upper_estimate": 0.370679039438635
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08582555807025471,
-    "producer_p95_seconds": 0.19110802300019714,
-    "gpu_update_seconds": 0.9531111942892092,
-    "producer_batches_per_second": 11.651540898591353,
-    "gpu_batches_per_second": 1.0491955251304739,
-    "producer_headroom": 11.10521406116597,
+    "producer_batch_seconds": 0.1477478066873772,
+    "producer_p95_seconds": 0.18877162500029954,
+    "gpu_update_seconds": 0.956727573340961,
+    "producer_batches_per_second": 6.768289982916104,
+    "gpu_batches_per_second": 1.04522962216708,
+    "producer_headroom": 6.475409651023258,
     "sustainable": true,
-    "buffer_seconds": 60.04600524022018,
+    "buffer_seconds": 60.27383712048054,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370645.90669178474,
-    "upload_bytes_per_second": 4280030.941561413,
+    "generation_bytes_per_second": 360187.8990597699,
+    "upload_bytes_per_second": 4267705.567515108,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.53198338093907,
+    "estimated_full_shards_per_hour": 77.28793839306661,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 1239114,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 2.040032099999735,
+      "bytes": 7666715,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -107,3 +107,4 @@
 {"step": 935700, "loss": 0.5434572696685791, "lr": 3e-05, "grad_norm": 0.32923772548568175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136960174468, "tokens_per_second": 153736.71112189864, "data_wait_s": 1.3414000932243653e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 935800, "loss": 0.5278041958808899, "lr": 3e-05, "grad_norm": 0.3256821274794918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136974846785, "tokens_per_second": 153273.4032538925, "data_wait_s": 1.2362001143628731e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 935900, "loss": 0.5134127736091614, "lr": 3e-05, "grad_norm": 0.33227880597816, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136989423644, "tokens_per_second": 152185.21245757845, "data_wait_s": 1.406699993822258e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 936000, "loss": 0.5576494932174683, "lr": 3e-05, "grad_norm": 0.4360252231470891, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137003943976, "tokens_per_second": 146671.57886310955, "data_wait_s": 1.644500116526615e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3b63aa32d1bcf820d1a2c3b4512dc9b08b7947e74312c72d6325bda4f746dc9
-size 232593
+oid sha256:d020090a6adc0e805592b3d804cdad6c959483db7028291f6ce679b29f730f24
+size 234711