Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75278.36745082504,
-      "required_pages_per_second_upper_estimate": 0.006524938099359509
+      "required_token_positions_per_second": 75366.5721772545,
+      "required_pages_per_second_upper_estimate": 0.006532583461493578
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9409.79593135313,
-      "required_pages_per_second_upper_estimate": 0.003753708286003323
+      "required_token_positions_per_second": 9420.821522156812,
+      "required_pages_per_second_upper_estimate": 0.0037581065590221842
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18819.59186270626,
-      "required_pages_per_second_upper_estimate": 0.004353581363932655
+      "required_token_positions_per_second": 18841.643044313623,
+      "required_pages_per_second_upper_estimate": 0.004358682516709978
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28229.387794059392,
-      "required_pages_per_second_upper_estimate": 0.0017190216085068626
+      "required_token_positions_per_second": 28262.464566470433,
+      "required_pages_per_second_upper_estimate": 0.0017210358104062821
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28229.387794059392,
-      "required_pages_per_second_upper_estimate": 0.002173176416333541
+      "required_token_positions_per_second": 28262.464566470433,
+      "required_pages_per_second_upper_estimate": 0.002175722757836109
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28229.387794059392,
-      "required_pages_per_second_upper_estimate": 0.35051451871884215
+      "required_token_positions_per_second": 28262.464566470433,
+      "required_pages_per_second_upper_estimate": 0.3509252215313512
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1202717069061805,
-    "producer_p95_seconds": 0.2674852730015118,
-    "gpu_update_seconds": 0.9794048741580667,
-    "producer_batches_per_second": 8.31450742426116,
-    "gpu_batches_per_second": 1.0210282043568935,
-    "producer_headroom": 8.143269097544813,
+    "producer_batch_seconds": 0.35389315090623086,
+    "producer_p95_seconds": 0.2798306269978639,
+    "gpu_update_seconds": 0.9782586346981427,
+    "producer_batches_per_second": 2.8257116517775294,
+    "gpu_batches_per_second": 1.0222245575256956,
+    "producer_headroom": 2.7642768225185192,
     "sustainable": true,
-    "buffer_seconds": 60.72310219780013,
+    "buffer_seconds": 60.652035351284844,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392625.68957685307,
-    "upload_bytes_per_second": 6341590.267958595,
+    "generation_bytes_per_second": 362857.6865959167,
+    "upload_bytes_per_second": 6324833.565360095,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.24833312491602,
+    "estimated_full_shards_per_hour": 77.86081264885068,
     "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": 1.0362921949999873,
-      "bytes": 5998979,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 0.7888650649983902,
       "bytes": 730238,
@@ -562,6 +556,12 @@
       "bytes": 12283477,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 1.1743630839991965,
+      "bytes": 5760431,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -161,3 +161,4 @@
 {"step": 56100, "loss": 0.6293352022767067, "lr": 0.0003, "grad_norm": 0.12303810071863236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8211177499, "tokens_per_second": 147823.8954392136, "data_wait_s": 1.2841002899222076e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 56200, "loss": 0.6209012195467949, "lr": 0.0003, "grad_norm": 0.1692370974729634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8225679998, "tokens_per_second": 148443.13043284757, "data_wait_s": 1.0531002772040665e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 56300, "loss": 0.6548167616128922, "lr": 0.0003, "grad_norm": 0.13857109300162795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8240281188, "tokens_per_second": 146028.79511833427, "data_wait_s": 8.115999662550166e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 56400, "loss": 0.6107414867728949, "lr": 0.0003, "grad_norm": 0.14392933913865694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8254997053, "tokens_per_second": 150781.31389081103, "data_wait_s": 1.1790001735789701e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3630660dcb1302bfc09217a52d291c0130200cb391e7a50d317e32c1c042f45
-size 347748
+oid sha256:7f35df10e401ed77705b5192b9b7a5c87fcb68ccc5e94679c0e6f639bc6ea051
+size 349866