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": 76199.33342228306,
-      "required_pages_per_second_upper_estimate": 0.005271646692966307
+      "required_token_positions_per_second": 76076.50108351244,
+      "required_pages_per_second_upper_estimate": 0.005263148866759863
     },
     {
       "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": 9524.916677785382,
-      "required_pages_per_second_upper_estimate": 0.003987883710110507
+      "required_token_positions_per_second": 9509.562635439055,
+      "required_pages_per_second_upper_estimate": 0.003981455293208964
     },
     {
       "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": 19049.833355570765,
-      "required_pages_per_second_upper_estimate": 0.0039675779100763545
+      "required_token_positions_per_second": 19019.12527087811,
+      "required_pages_per_second_upper_estimate": 0.003961182225861526
     },
     {
       "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": 28574.75003335615,
-      "required_pages_per_second_upper_estimate": 0.001713012633382724
+      "required_token_positions_per_second": 28528.68790631717,
+      "required_pages_per_second_upper_estimate": 0.0017102512791995325
     },
     {
       "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": 28574.75003335615,
-      "required_pages_per_second_upper_estimate": 0.00222156112805106
+      "required_token_positions_per_second": 28528.68790631717,
+      "required_pages_per_second_upper_estimate": 0.002217980000279664
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28574.75003335615,
-      "required_pages_per_second_upper_estimate": 0.36652621225171755
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28528.68790631717,
+      "required_pages_per_second_upper_estimate": 0.3659353767437202
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.19577828164838706,
-    "producer_p95_seconds": 0.25866901500194217,
-    "gpu_update_seconds": 0.9675675191462978,
-    "producer_batches_per_second": 5.1078188631565125,
-    "gpu_batches_per_second": 1.0335196047944208,
-    "producer_headroom": 4.9421596256730105,
+    "producer_batch_seconds": 0.09052958598425676,
+    "producer_p95_seconds": 0.27030788699994446,
+    "gpu_update_seconds": 0.9691297437439401,
+    "producer_batches_per_second": 11.046112595432632,
+    "gpu_batches_per_second": 1.0318535845745502,
+    "producer_headroom": 10.705116268978335,
     "sustainable": true,
-    "buffer_seconds": 60.95675370621676,
+    "buffer_seconds": 60.086044112124284,
     "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": 505033.97971716744,
-    "upload_bytes_per_second": 3935413.792238827,
+    "generation_bytes_per_second": 510082.6641006888,
+    "upload_bytes_per_second": 3982873.927649975,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.36853545795695,
+    "estimated_full_shards_per_hour": 109.45186559930323,
     "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": 1.8639257160029956,
-      "bytes": 7235620,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.8414086459997634,
       "bytes": 5860537,
@@ -562,6 +556,12 @@
       "bytes": 7597965,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.836660891000065,
+      "bytes": 12233872,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -233,3 +233,4 @@
 {"step": 823300, "loss": 0.5318233370780945, "lr": 3e-05, "grad_norm": 0.2707454232859008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120508057249, "tokens_per_second": 152149.08682513738, "data_wait_s": 2.482100171619095e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823400, "loss": 0.5219301581382751, "lr": 3e-05, "grad_norm": 0.28984945960498276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120522630604, "tokens_per_second": 150153.22537021586, "data_wait_s": 1.5391000488307327e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823500, "loss": 0.5565692186355591, "lr": 3e-05, "grad_norm": 0.27770056215623146, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120537270535, "tokens_per_second": 151278.89193235245, "data_wait_s": 1.1182997695868835e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 823600, "loss": 0.5208242535591125, "lr": 3e-05, "grad_norm": 0.2910052313811568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120551893798, "tokens_per_second": 151255.03495848106, "data_wait_s": 1.5976998838596046e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6ce9c2988dbb75c3d2da178653e375b335334cb4e6903523a4587f50bea725f5
-size 501027
+oid sha256:65cf2771f9719a79b586f3ef542fbaa57f30a4d56054bc1e920709774537f16c
+size 503145