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": 77400.48435629702,
-      "required_pages_per_second_upper_estimate": 0.00535474510163554
+      "required_token_positions_per_second": 76948.83129662581,
+      "required_pages_per_second_upper_estimate": 0.005323498694988004
     },
     {
       "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": 9675.060544537127,
-      "required_pages_per_second_upper_estimate": 0.004050745811758991
+      "required_token_positions_per_second": 9618.603912078226,
+      "required_pages_per_second_upper_estimate": 0.004027108598696998
     },
     {
       "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": 19350.121089074255,
-      "required_pages_per_second_upper_estimate": 0.004030119925845061
+      "required_token_positions_per_second": 19237.207824156452,
+      "required_pages_per_second_upper_estimate": 0.0040066030704857215
     },
     {
       "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": 29025.18163361138,
-      "required_pages_per_second_upper_estimate": 0.0017400153200487945
+      "required_token_positions_per_second": 28855.81173623468,
+      "required_pages_per_second_upper_estimate": 0.0017298618533139202
     },
     {
       "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": 29025.18163361138,
-      "required_pages_per_second_upper_estimate": 0.0022565802037316784
+      "required_token_positions_per_second": 28855.81173623468,
+      "required_pages_per_second_upper_estimate": 0.002243412439190087
     },
     {
       "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": 29025.18163361138,
-      "required_pages_per_second_upper_estimate": 0.37230386518402
+      "required_token_positions_per_second": 28855.81173623468,
+      "required_pages_per_second_upper_estimate": 0.37013136999569884
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13981808507804772,
-    "producer_p95_seconds": 0.3362951390008675,
-    "gpu_update_seconds": 0.9525521786221454,
-    "producer_batches_per_second": 7.152150592262732,
-    "gpu_batches_per_second": 1.0498112570027265,
-    "producer_headroom": 6.812796628493532,
+    "producer_batch_seconds": 0.16316515332010795,
+    "producer_p95_seconds": 0.32672257600643206,
+    "gpu_update_seconds": 0.9581432070851083,
+    "producer_batches_per_second": 6.128759601249755,
+    "gpu_batches_per_second": 1.0436853203209882,
+    "producer_headroom": 5.8722293797950895,
     "sustainable": true,
-    "buffer_seconds": 60.01078725319516,
+    "buffer_seconds": 60.36302204636182,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 315923.13584238605,
-    "upload_bytes_per_second": 5498054.990331841,
+    "generation_bytes_per_second": 320497.9659336169,
+    "upload_bytes_per_second": 5445753.153991307,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.78975063756643,
+    "estimated_full_shards_per_hour": 68.77140267855053,
     "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": 0.9623371359994053,
-      "bytes": 2372191,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.2589823859998432,
       "bytes": 12076255,
@@ -562,6 +556,12 @@
       "bytes": 5905666,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 2.588622392999241,
+      "bytes": 7185173,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -334,3 +334,4 @@
 {"step": 473600, "loss": 0.557610273361206, "lr": 0.0003, "grad_norm": 0.19698444025243064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69320354325, "tokens_per_second": 153501.88476630833, "data_wait_s": 1.2992000847589225e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 473700, "loss": 0.5600315928459167, "lr": 0.0003, "grad_norm": 0.17488606683979363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69334970703, "tokens_per_second": 125777.29565164473, "data_wait_s": 9.38499579206109e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 473800, "loss": 0.6174167394638062, "lr": 0.0003, "grad_norm": 0.17022402488085814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69349458590, "tokens_per_second": 150756.6135311059, "data_wait_s": 1.1961994459852576e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 473900, "loss": 0.5944867134094238, "lr": 0.0003, "grad_norm": 0.1693777827876962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69364032372, "tokens_per_second": 151665.13956717157, "data_wait_s": 9.452000085730106e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:12ce8816c9f275d0c28d16545989cb17f08036a0b89153e2c92bc8d7b3e67176
-size 716511
+oid sha256:d1a4c1a8c48f9d3d95ffbad87a94d4afa2ee0ce6a2b5a9d0c0a1435355fbe3bc
+size 718629