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": 77317.05426615606,
-      "required_pages_per_second_upper_estimate": 0.005348973214415109
+      "required_token_positions_per_second": 77512.26140783782,
+      "required_pages_per_second_upper_estimate": 0.005362478097419626
     },
     {
       "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": 9664.631783269508,
-      "required_pages_per_second_upper_estimate": 0.004046379507193538
+      "required_token_positions_per_second": 9689.032675979728,
+      "required_pages_per_second_upper_estimate": 0.004056595651422725
     },
     {
       "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": 19329.263566539015,
-      "required_pages_per_second_upper_estimate": 0.004025775853950829
+      "required_token_positions_per_second": 19378.065351959456,
+      "required_pages_per_second_upper_estimate": 0.004035939978864279
     },
     {
       "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": 28993.89534980852,
-      "required_pages_per_second_upper_estimate": 0.0017381397551062025
+      "required_token_positions_per_second": 29067.098027939184,
+      "required_pages_per_second_upper_estimate": 0.001742528143886118
     },
     {
       "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": 28993.89534980852,
-      "required_pages_per_second_upper_estimate": 0.0022541478327798253
+      "required_token_positions_per_second": 29067.098027939184,
+      "required_pages_per_second_upper_estimate": 0.0022598390190199315
     },
     {
       "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": 28993.89534980852,
-      "required_pages_per_second_upper_estimate": 0.3719025583279912
+      "required_token_positions_per_second": 29067.098027939184,
+      "required_pages_per_second_upper_estimate": 0.3728415236841393
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,17 +155,17 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04772446601562841,
-    "producer_p95_seconds": 0.12866096200014,
-    "gpu_update_seconds": 0.9535800438826717,
-    "producer_batches_per_second": 20.95361317762106,
-    "gpu_batches_per_second": 1.0486796639832363,
-    "producer_headroom": 19.98094737341642,
+    "producer_batch_seconds": 0.050092538976567624,
+    "producer_p95_seconds": 0.13427479699998912,
+    "gpu_update_seconds": 0.9511785446701575,
+    "producer_batches_per_second": 19.963052790512013,
+    "gpu_batches_per_second": 1.0513273302929393,
+    "producer_headroom": 18.98842750045274,
     "sustainable": true,
-    "buffer_seconds": 60.075542764608315,
+    "buffer_seconds": 60.87542685889008,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 67,

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -10,3 +10,4 @@
 {"step": 686000, "loss": 0.5892570614814758, "lr": 8e-05, "grad_norm": 0.21885647328773225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100412502302, "tokens_per_second": 152886.282932156, "data_wait_s": 1.1249999943174771e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686100, "loss": 0.5920264720916748, "lr": 8e-05, "grad_norm": 0.2095812736589121, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100427191750, "tokens_per_second": 153705.14991769212, "data_wait_s": 1.0631999884935794e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 686200, "loss": 0.5173473358154297, "lr": 8e-05, "grad_norm": 0.21569271361113784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100441963613, "tokens_per_second": 155056.08306677538, "data_wait_s": 1.6467000023112632e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 686300, "loss": 0.5384304523468018, "lr": 8e-05, "grad_norm": 0.19933378543151312, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100456651841, "tokens_per_second": 154290.50755349858, "data_wait_s": 1.608100001249113e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c6105cd8efa73dacf38995acd57bad5eaadf923d49da0e828f8d8b02d90f637
-size 25581
+oid sha256:8cce6475cf55cfc4b503aa191e3f6d7e3121f9981125b5ac1c08dac532dcd054
+size 27699