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": 74952.30606969121,
-      "required_pages_per_second_upper_estimate": 0.006496675925237844
+      "required_token_positions_per_second": 74891.70315754368,
+      "required_pages_per_second_upper_estimate": 0.006491423018409573
     },
     {
       "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": 9369.038258711402,
-      "required_pages_per_second_upper_estimate": 0.003737449441005027
+      "required_token_positions_per_second": 9361.46289469296,
+      "required_pages_per_second_upper_estimate": 0.0037344275150362856
     },
     {
       "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": 18738.076517422804,
-      "required_pages_per_second_upper_estimate": 0.004334724223422406
+      "required_token_positions_per_second": 18722.92578938592,
+      "required_pages_per_second_upper_estimate": 0.00433121936913478
     },
     {
       "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": 28107.114776134207,
-      "required_pages_per_second_upper_estimate": 0.0017115758232321648
+      "required_token_positions_per_second": 28084.38868407888,
+      "required_pages_per_second_upper_estimate": 0.0017101919234605856
     },
     {
       "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": 28107.114776134207,
-      "required_pages_per_second_upper_estimate": 0.0021637635009402836
+      "required_token_positions_per_second": 28084.38868407888,
+      "required_pages_per_second_upper_estimate": 0.002162013983464015
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28107.114776134207,
-      "required_pages_per_second_upper_estimate": 0.34899629705767793
+      "required_token_positions_per_second": 28084.38868407888,
+      "required_pages_per_second_upper_estimate": 0.3487141150536881
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.6708330393436768,
-    "producer_p95_seconds": 0.2904391099982604,
-    "gpu_update_seconds": 0.9836655316708621,
-    "producer_batches_per_second": 1.490683883099095,
-    "gpu_batches_per_second": 1.0166057138358726,
-    "producer_headroom": 1.4663343544218566,
+    "producer_batch_seconds": 0.0883145759453896,
+    "producer_p95_seconds": 0.2765440340008354,
+    "gpu_update_seconds": 0.984461520989906,
+    "producer_batches_per_second": 11.323159164783423,
+    "gpu_batches_per_second": 1.015783734233177,
+    "producer_headroom": 11.147214493773483,
     "sustainable": true,
-    "buffer_seconds": 60.00359743192259,
+    "buffer_seconds": 60.05215278038427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318744.74860939826,
+    "generation_bytes_per_second": 318938.4005100719,
     "upload_bytes_per_second": 5747651.376155982,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.39520305358373,
+    "estimated_full_shards_per_hour": 68.43675624348276,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -210,3 +210,4 @@
 {"step": 376000, "loss": 0.6168783605098724, "lr": 0.0003, "grad_norm": 0.1514677382010011, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55035484909, "tokens_per_second": 149685.17553079486, "data_wait_s": 1.2767002772307023e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 376100, "loss": 0.6699020639061928, "lr": 0.0003, "grad_norm": 0.16973057572364986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55049925175, "tokens_per_second": 125615.81850441806, "data_wait_s": 1.5583002095809206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 376200, "loss": 0.6044797673821449, "lr": 0.0003, "grad_norm": 0.1593471372290677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55064479747, "tokens_per_second": 148061.10062921714, "data_wait_s": 8.700000762473792e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 376300, "loss": 0.636867169290781, "lr": 0.0003, "grad_norm": 0.16422846199396238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55079089404, "tokens_per_second": 148924.64785953984, "data_wait_s": 1.2706997949862853e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1122e9be422d6b6ece95bf155ad519f2f07f777b5032f10a8c42ee022438336a
-size 452313
+oid sha256:939356be143cfb49a210a4a8e29d9439d8bec996d5c1c29cfdb7d17c6ed99323
+size 454431