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": 75333.49158158887,
-      "required_pages_per_second_upper_estimate": 0.006529716119303286
+      "required_token_positions_per_second": 75064.75402666364,
+      "required_pages_per_second_upper_estimate": 0.006506422629151473
     },
     {
       "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": 9416.686447698608,
-      "required_pages_per_second_upper_estimate": 0.0037564570159959346
+      "required_token_positions_per_second": 9383.094253332954,
+      "required_pages_per_second_upper_estimate": 0.0037430565874154117
     },
     {
       "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": 18833.372895397217,
-      "required_pages_per_second_upper_estimate": 0.004356769363307817
+      "required_token_positions_per_second": 18766.18850666591,
+      "required_pages_per_second_upper_estimate": 0.004341227437379686
     },
     {
       "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": 28250.059343095825,
-      "required_pages_per_second_upper_estimate": 0.0017202803973879455
+      "required_token_positions_per_second": 28149.282759998863,
+      "required_pages_per_second_upper_estimate": 0.0017141436322112092
     },
     {
       "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": 28250.059343095825,
-      "required_pages_per_second_upper_estimate": 0.0021747677694008757
+      "required_token_positions_per_second": 28149.282759998863,
+      "required_pages_per_second_upper_estimate": 0.002167009709066635
     },
     {
       "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": 28250.059343095825,
-      "required_pages_per_second_upper_estimate": 0.35077119017465047
+      "required_token_positions_per_second": 28149.282759998863,
+      "required_pages_per_second_upper_estimate": 0.3495198822901134
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.053614596093765954,
-    "producer_p95_seconds": 0.1821551310003997,
-    "gpu_update_seconds": 0.9786882096145768,
-    "producer_batches_per_second": 18.651637293902418,
-    "gpu_batches_per_second": 1.0217758732311857,
-    "producer_headroom": 18.254137509549828,
+    "producer_batch_seconds": 0.08264098267979136,
+    "producer_p95_seconds": 0.19437852100054442,
+    "gpu_update_seconds": 0.9821919881841109,
+    "producer_batches_per_second": 12.100533749395206,
+    "gpu_batches_per_second": 1.0181308868633847,
+    "producer_headroom": 11.885047301407411,
     "sustainable": true,
-    "buffer_seconds": 60.67866899610376,
+    "buffer_seconds": 60.89590326741487,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 264899.4268516722,
-    "upload_bytes_per_second": 5579095.736845195,
+    "generation_bytes_per_second": 262814.54113664053,
+    "upload_bytes_per_second": 5484321.6593935685,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.84125045931457,
+    "estimated_full_shards_per_hour": 56.393882518524286,
     "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": [
@@ -394,6 +394,12 @@
       "bytes": 17104134,
       "pages": 27,
       "shards": 5
+    },
+    {
+      "seconds": 0.9077872060006484,
+      "bytes": 916446,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -43,3 +43,4 @@
 {"step": 284500, "loss": 0.5872170403599739, "lr": 0.0003, "grad_norm": 0.16318429725297312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41642989966, "tokens_per_second": 150177.27312181768, "data_wait_s": 1.2517999493866228e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 284600, "loss": 0.6178310494869947, "lr": 0.0003, "grad_norm": 0.1553548230141656, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41657633602, "tokens_per_second": 96489.93389584642, "data_wait_s": 9.039000360644422e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 284700, "loss": 0.5720347091555595, "lr": 0.0003, "grad_norm": 0.14328422273474825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41672295409, "tokens_per_second": 150428.9152204235, "data_wait_s": 1.4390001524589024e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 284800, "loss": 0.6524676568806171, "lr": 0.0003, "grad_norm": 0.1489935903173233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 41686821846, "tokens_per_second": 148659.5400859823, "data_wait_s": 1.6076000974862836e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e8c1716f70c007ae7a4ba53909fc89827eaf312bbb911b69c43ec065cd445003
-size 95475
+oid sha256:c7ab141263e53261fa3ada4370b124a0487894b5ee3dd35ef3cf5254ba0d4cb3
+size 97593