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": 77007.077432016,
-      "required_pages_per_second_upper_estimate": 0.005327528297783952
+      "required_token_positions_per_second": 77556.54585002972,
+      "required_pages_per_second_upper_estimate": 0.005365541797884524
     },
     {
       "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": 9625.884679002,
-      "required_pages_per_second_upper_estimate": 0.00403015690376828
+      "required_token_positions_per_second": 9694.568231253716,
+      "required_pages_per_second_upper_estimate": 0.004058913272820405
     },
     {
       "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": 19251.769358004,
-      "required_pages_per_second_upper_estimate": 0.004009635853972695
+      "required_token_positions_per_second": 19389.13646250743,
+      "required_pages_per_second_upper_estimate": 0.0040382457992266035
     },
     {
       "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": 28877.654037005996,
-      "required_pages_per_second_upper_estimate": 0.0017311712658938986
+      "required_token_positions_per_second": 29083.704693761145,
+      "required_pages_per_second_upper_estimate": 0.001743523688145234
     },
     {
       "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": 28877.654037005996,
-      "required_pages_per_second_upper_estimate": 0.002245110582000929
+      "required_token_positions_per_second": 29083.704693761145,
+      "required_pages_per_second_upper_estimate": 0.0022611301142425855
     },
     {
       "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": 28877.654037005996,
-      "required_pages_per_second_upper_estimate": 0.37041153957755796
+      "required_token_positions_per_second": 29083.704693761145,
+      "required_pages_per_second_upper_estimate": 0.3730545361624549
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 704566,
-    "producer_batch_seconds": 0.44668049865606463,
-    "producer_p95_seconds": 0.2637138069985667,
-    "gpu_update_seconds": 0.9574184926715229,
-    "producer_batches_per_second": 2.2387366428772184,
-    "gpu_batches_per_second": 1.0444753340930988,
-    "producer_headroom": 2.143407862112012,
+    "producer_batch_seconds": 0.09164676815635175,
+    "producer_p95_seconds": 0.23588989000199945,
+    "gpu_update_seconds": 0.9506354259583332,
+    "producer_batches_per_second": 10.911459510432207,
+    "gpu_batches_per_second": 1.0519279764815228,
+    "producer_headroom": 10.372819959526828,
     "sustainable": true,
-    "buffer_seconds": 60.317365038305944,
+    "buffer_seconds": 60.84066726133332,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357039.18204732693,
-    "upload_bytes_per_second": 5557295.794033938,
+    "generation_bytes_per_second": 357102.1713049466,
+    "upload_bytes_per_second": 5592660.902341825,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.61229702057702,
+    "estimated_full_shards_per_hour": 76.62581304894732,
     "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": [
-    {
-      "seconds": 1.2941537009992317,
-      "bytes": 6225550,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2335469089994149,
       "bytes": 11703737,
@@ -562,6 +556,12 @@
       "bytes": 5853547,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 0.9646089939997182,
+      "bytes": 7151890,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -166,3 +166,4 @@
 {"step": 501600, "loss": 0.5874972939491272, "lr": 0.0003, "grad_norm": 0.14712835035756988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73417566943, "tokens_per_second": 151537.9510395918, "data_wait_s": 1.1756997992051765e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 501700, "loss": 0.6177993416786194, "lr": 0.0003, "grad_norm": 0.16255062856013408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73432009424, "tokens_per_second": 149963.55805931397, "data_wait_s": 1.1176001862622797e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 501800, "loss": 0.5342846512794495, "lr": 0.0003, "grad_norm": 0.15386128369866137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73446603551, "tokens_per_second": 152406.28316549293, "data_wait_s": 1.8790000467561185e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 501900, "loss": 0.5525649785995483, "lr": 0.0003, "grad_norm": 0.1397344511084202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73461282059, "tokens_per_second": 153024.28076948592, "data_wait_s": 1.2282998795853928e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c4043df944bb18db753644d76a922eebc1f8fbc1275325cee851335e9dfe71e7
-size 358338
+oid sha256:8dbecdc6fa8186ffb5740489e7e93b7666f0a1e316813847245b985c32d6f8a9
+size 360456