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": 75988.61059300757,
-      "required_pages_per_second_upper_estimate": 0.005257068398692701
+      "required_token_positions_per_second": 76235.19121332189,
+      "required_pages_per_second_upper_estimate": 0.005274127418151911
     },
     {
       "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": 9498.576324125946,
-      "required_pages_per_second_upper_estimate": 0.003976855554082434
+      "required_token_positions_per_second": 9529.398901665236,
+      "required_pages_per_second_upper_estimate": 0.003989760323649524
     },
     {
       "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": 18997.152648251893,
-      "required_pages_per_second_upper_estimate": 0.00395660590802551
+      "required_token_positions_per_second": 19058.79780333047,
+      "required_pages_per_second_upper_estimate": 0.003969444968136316
     },
     {
       "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": 28495.72897237784,
-      "required_pages_per_second_upper_estimate": 0.0017082754414352481
+      "required_token_positions_per_second": 28588.19670499571,
+      "required_pages_per_second_upper_estimate": 0.0017138187408156367
     },
     {
       "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": 28495.72897237784,
-      "required_pages_per_second_upper_estimate": 0.0022154175881369106
+      "required_token_positions_per_second": 28588.19670499571,
+      "required_pages_per_second_upper_estimate": 0.002222606547625378
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28495.72897237784,
-      "required_pages_per_second_upper_estimate": 0.3655126149276926
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28588.19670499571,
+      "required_pages_per_second_upper_estimate": 0.36669869171759867
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08119746129684557,
-    "producer_p95_seconds": 0.1632093359994542,
-    "gpu_update_seconds": 0.9702506655225567,
-    "producer_batches_per_second": 12.315655982693253,
-    "gpu_batches_per_second": 1.0306614935032494,
-    "producer_headroom": 11.949273413554986,
+    "producer_batch_seconds": 0.24396757208588582,
+    "producer_p95_seconds": 0.17687704699983442,
+    "gpu_update_seconds": 0.9671124165438735,
+    "producer_batches_per_second": 4.09890540554284,
+    "gpu_batches_per_second": 1.0340059572119396,
+    "producer_headroom": 3.9641023119392824,
     "sustainable": true,
-    "buffer_seconds": 60.155541262398515,
+    "buffer_seconds": 60.928082242264026,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 385277.1020865466,
+    "generation_bytes_per_second": 382409.18981197616,
     "upload_bytes_per_second": 3731167.4925377467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.67149731585788,
+    "estimated_full_shards_per_hour": 82.05611010331596,
     "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": [

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

@@ -54,3 +54,4 @@
 {"step": 805400, "loss": 0.5579742193222046, "lr": 3e-05, "grad_norm": 0.28295588344758865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117887288118, "tokens_per_second": 145349.65456967396, "data_wait_s": 1.2562999472720549e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 805500, "loss": 0.5162538886070251, "lr": 3e-05, "grad_norm": 0.2728224342692626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117902043224, "tokens_per_second": 152878.94299546952, "data_wait_s": 8.201000127883162e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 805600, "loss": 0.5071325898170471, "lr": 3e-05, "grad_norm": 0.27077951718871107, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117916661451, "tokens_per_second": 151694.02694097103, "data_wait_s": 2.65179996858933e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 805700, "loss": 0.5595177412033081, "lr": 3e-05, "grad_norm": 0.27615446483562395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117931176621, "tokens_per_second": 150598.8711746715, "data_wait_s": 1.2598000466823578e-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 sha256:6b913f4fd8fcf41fbfcb8ee8c459a41a291c4942e7ff956255293dbbcfa2276e
-size 119556
+oid sha256:da6086aea01caf20ba3ee4237fe2830082ca2cbad902333f801db3310ed443f5
+size 121674