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": 77362.25690637971,
-      "required_pages_per_second_upper_estimate": 0.005352100437950357
+      "required_token_positions_per_second": 77176.33676402176,
+      "required_pages_per_second_upper_estimate": 0.005339238051107869
     },
     {
       "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": 9670.282113297464,
-      "required_pages_per_second_upper_estimate": 0.0040487451823839355
+      "required_token_positions_per_second": 9647.04209550272,
+      "required_pages_per_second_upper_estimate": 0.004039015072240034
     },
     {
       "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": 19340.56422659493,
-      "required_pages_per_second_upper_estimate": 0.004028129483422029
+      "required_token_positions_per_second": 19294.08419100544,
+      "required_pages_per_second_upper_estimate": 0.004018448917769706
     },
     {
       "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": 29010.846339892392,
-      "required_pages_per_second_upper_estimate": 0.0017391559410790674
+      "required_token_positions_per_second": 28941.126286508163,
+      "required_pages_per_second_upper_estimate": 0.0017349763303350402
     },
     {
       "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": 29010.846339892392,
-      "required_pages_per_second_upper_estimate": 0.0022554656976993205
+      "required_token_positions_per_second": 28941.126286508163,
+      "required_pages_per_second_upper_estimate": 0.002250045270214802
     },
     {
       "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": 29010.846339892392,
-      "required_pages_per_second_upper_estimate": 0.372119987428232
+      "required_token_positions_per_second": 28941.126286508163,
+      "required_pages_per_second_upper_estimate": 0.37122569344298
     }
   ],
   "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": 0.10334270342193008,
-    "producer_p95_seconds": 0.3099736000003759,
-    "gpu_update_seconds": 0.9530228686221276,
-    "producer_batches_per_second": 9.676541902694144,
-    "gpu_batches_per_second": 1.0492927640296728,
-    "producer_headroom": 9.221965722447795,
+    "producer_batch_seconds": 0.12006736707803611,
+    "producer_p95_seconds": 0.31453827299992554,
+    "gpu_update_seconds": 0.9553187296960521,
+    "producer_batches_per_second": 8.328657688896135,
+    "gpu_batches_per_second": 1.0467710607099305,
+    "producer_headroom": 7.956522683429512,
     "sustainable": true,
-    "buffer_seconds": 60.04044072319404,
+    "buffer_seconds": 60.185079970851284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382166.92027744383,
+    "generation_bytes_per_second": 382589.9067359196,
     "upload_bytes_per_second": 4904583.399781482,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.00412470095145,
+    "estimated_full_shards_per_hour": 82.09488774831954,
     "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": [

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

@@ -308,3 +308,4 @@
 {"step": 955800, "loss": 0.5293927192687988, "lr": 3e-05, "grad_norm": 0.34714447391276426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139902951883, "tokens_per_second": 153932.88179856248, "data_wait_s": 1.3213997590355575e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 955900, "loss": 0.5105063319206238, "lr": 3e-05, "grad_norm": 0.3532509704132093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139917613706, "tokens_per_second": 152705.56575056832, "data_wait_s": 1.0668001777958125e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 956000, "loss": 0.5017450451850891, "lr": 3e-05, "grad_norm": 0.3368989881529334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139932144721, "tokens_per_second": 151141.53548767915, "data_wait_s": 1.2212000001454726e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 956100, "loss": 0.5179443955421448, "lr": 3e-05, "grad_norm": 0.35472210072890226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139946608043, "tokens_per_second": 150216.62423743444, "data_wait_s": 1.1257001460762694e-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 sha256:0d45a50f3365a58d6ea004a7d73f9c583053c44a4985589c0925fccd9d130b17
-size 661443
+oid sha256:aeb6b0e35c6d22d682206063b917d471b5b392d8cb50c9d8d82f02af231e76f7
+size 663561