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": 75962.00820263596,
-      "required_pages_per_second_upper_estimate": 0.006584194347054835
+      "required_token_positions_per_second": 75877.15650200497,
+      "required_pages_per_second_upper_estimate": 0.00657683961669881
     },
     {
       "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": 9495.251025329495,
-      "required_pages_per_second_upper_estimate": 0.003787797600658008
+      "required_token_positions_per_second": 9484.644562750622,
+      "required_pages_per_second_upper_estimate": 0.0037835665241545482
     },
     {
       "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": 18990.50205065899,
-      "required_pages_per_second_upper_estimate": 0.004393118427998945
+      "required_token_positions_per_second": 18969.289125501244,
+      "required_pages_per_second_upper_estimate": 0.004388211191098432
     },
     {
       "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": 28485.753075988483,
-      "required_pages_per_second_upper_estimate": 0.0017346329091316604
+      "required_token_positions_per_second": 28453.933688251866,
+      "required_pages_per_second_upper_estimate": 0.0017326952753619258
     },
     {
       "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": 28485.753075988483,
-      "required_pages_per_second_upper_estimate": 0.0021929121254009635
+      "required_token_positions_per_second": 28453.933688251866,
+      "required_pages_per_second_upper_estimate": 0.002190462580851296
     },
     {
       "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": 28485.753075988483,
-      "required_pages_per_second_upper_estimate": 0.3536977175209963
+      "required_token_positions_per_second": 28453.933688251866,
+      "required_pages_per_second_upper_estimate": 0.35330262721794786
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.06604437228899229,
-    "producer_p95_seconds": 0.17501575400001457,
-    "gpu_update_seconds": 0.9705904536294443,
-    "producer_batches_per_second": 15.141335519463652,
-    "gpu_batches_per_second": 1.0303006754914816,
-    "producer_headroom": 14.696035710391845,
+    "producer_batch_seconds": 0.05315762628120524,
+    "producer_p95_seconds": 0.17704411900012929,
+    "gpu_update_seconds": 0.9716758428875997,
+    "producer_batches_per_second": 18.81197619152469,
+    "gpu_batches_per_second": 1.0291498006456838,
+    "producer_headroom": 18.27914282228121,
     "sustainable": true,
-    "buffer_seconds": 60.176608125025545,
+    "buffer_seconds": 60.24390225903118,
     "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": 350497.3563680549,
-    "upload_bytes_per_second": 5442543.546901493,
+    "generation_bytes_per_second": 350214.6096440383,
+    "upload_bytes_per_second": 5315003.418443578,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.20857351571308,
+    "estimated_full_shards_per_hour": 75.14790265074598,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 6008702,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.7417863239998042,
+      "bytes": 879449,
+      "pages": 25,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -49,3 +49,4 @@
 {"step": 359900, "loss": 0.589072197675705, "lr": 0.0003, "grad_norm": 0.1709890691884968, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52679165005, "tokens_per_second": 148789.96262994417, "data_wait_s": 8.371000149054453e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 360000, "loss": 0.6037846226245165, "lr": 0.0003, "grad_norm": 0.14638077382152287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52693835160, "tokens_per_second": 149850.34213180692, "data_wait_s": 9.698000212665647e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 360100, "loss": 0.6022424828261137, "lr": 0.0003, "grad_norm": 0.18071048781891982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52708490102, "tokens_per_second": 127690.04584187643, "data_wait_s": 1.3590999515145086e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 360200, "loss": 0.5641546994447708, "lr": 0.0003, "grad_norm": 0.1564444480186743, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52723247920, "tokens_per_second": 150561.82350302662, "data_wait_s": 1.1824000466731377e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2545948532812d726755c01d62362e65ba0bb3f151dbc004f83f296181165ca8
-size 108966
+oid sha256:a656288bf4018782676e90ae2ca38735a706294be7bb10504dcb91628d4a19a7
+size 111084