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": 77259.17846097716,
-      "required_pages_per_second_upper_estimate": 0.005344969231922463
+      "required_token_positions_per_second": 77000.16840730303,
+      "required_pages_per_second_upper_estimate": 0.005327050315423163
     },
     {
       "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": 9657.397307622145,
-      "required_pages_per_second_upper_estimate": 0.004043350583313018
+      "required_token_positions_per_second": 9625.02105091288,
+      "required_pages_per_second_upper_estimate": 0.00402979532072197
     },
     {
       "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": 19314.79461524429,
-      "required_pages_per_second_upper_estimate": 0.004022762352968043
+      "required_token_positions_per_second": 19250.04210182576,
+      "required_pages_per_second_upper_estimate": 0.004009276112061575
     },
     {
       "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": 28972.191922866434,
-      "required_pages_per_second_upper_estimate": 0.0017368386678002384
+      "required_token_positions_per_second": 28875.063152738636,
+      "required_pages_per_second_upper_estimate": 0.0017310159463381216
     },
     {
       "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": 28972.191922866434,
-      "required_pages_per_second_upper_estimate": 0.002252460486798364
+      "required_token_positions_per_second": 28875.063152738636,
+      "required_pages_per_second_upper_estimate": 0.002244909152145237
     },
     {
       "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": 28972.191922866434,
-      "required_pages_per_second_upper_estimate": 0.37162417007050236
+      "required_token_positions_per_second": 28875.063152738636,
+      "required_pages_per_second_upper_estimate": 0.370378306496051
     }
   ],
   "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": 0.4321962123983951,
-    "producer_p95_seconds": 0.2989799379975011,
-    "gpu_update_seconds": 0.9542943824757246,
-    "producer_batches_per_second": 2.3137639139655573,
-    "gpu_batches_per_second": 1.0478946731360834,
-    "producer_headroom": 2.208011905472377,
+    "producer_batch_seconds": 0.137767387289216,
+    "producer_p95_seconds": 0.2926373079972109,
+    "gpu_update_seconds": 0.957504399341123,
+    "producer_batches_per_second": 7.25861192315924,
+    "gpu_batches_per_second": 1.044381624448012,
+    "producer_headroom": 6.950152849534901,
     "sustainable": true,
-    "buffer_seconds": 60.12054609597065,
+    "buffer_seconds": 60.322777158490744,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298357.30461501505,
-    "upload_bytes_per_second": 5776436.2947026,
+    "generation_bytes_per_second": 323623.7453334474,
+    "upload_bytes_per_second": 5758397.724492029,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.02053216779555,
+    "estimated_full_shards_per_hour": 69.4421221733338,
     "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": 0.9431873399989854,
-      "bytes": 5431625,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.3104591960000107,
       "bytes": 11823249,
@@ -562,6 +556,12 @@
       "bytes": 10000231,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.287032372001704,
+      "bytes": 6056411,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -317,3 +317,4 @@
 {"step": 428800, "loss": 0.5621210336685181, "lr": 0.0003, "grad_norm": 0.15101313219493362, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62764514705, "tokens_per_second": 150946.41698847347, "data_wait_s": 1.3472999853547662e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 428900, "loss": 0.5718140006065369, "lr": 0.0003, "grad_norm": 0.1473230188598794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62779181864, "tokens_per_second": 151555.54729809656, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 429000, "loss": 0.5638366341590881, "lr": 0.0003, "grad_norm": 0.15660331808988368, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62793871506, "tokens_per_second": 152104.7180156272, "data_wait_s": 8.949999028118327e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 429100, "loss": 0.6631481051445007, "lr": 0.0003, "grad_norm": 0.17248152837054945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62808446446, "tokens_per_second": 150982.356583782, "data_wait_s": 2.0313000277383253e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f84a63cd64159d5c77e8f37b1cf691bc1332232a7d7e371b11855f83d1c6b3f3
-size 680505
+oid sha256:e16b12807b1ccc2880371ffd7da8c05fe039d1e8779bdc70472241dd4395209c
+size 682623