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": 76958.28231469195,
-      "required_pages_per_second_upper_estimate": 0.005324152538347203
+      "required_token_positions_per_second": 77001.70021049575,
+      "required_pages_per_second_upper_estimate": 0.005327156289122308
     },
     {
       "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": 9619.785289336494,
-      "required_pages_per_second_upper_estimate": 0.004027603216685072
+      "required_token_positions_per_second": 9625.212526311969,
+      "required_pages_per_second_upper_estimate": 0.0040298754874731075
     },
     {
       "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": 19239.57057867299,
-      "required_pages_per_second_upper_estimate": 0.004007095169941489
+      "required_token_positions_per_second": 19250.425052623938,
+      "required_pages_per_second_upper_estimate": 0.004009355870613744
     },
     {
       "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": 28859.35586800948,
-      "required_pages_per_second_upper_estimate": 0.0017300743186022438
+      "required_token_positions_per_second": 28875.637578935904,
+      "required_pages_per_second_upper_estimate": 0.0017310503823115499
     },
     {
       "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": 28859.35586800948,
-      "required_pages_per_second_upper_estimate": 0.002243687979846589
+      "required_token_positions_per_second": 28875.637578935904,
+      "required_pages_per_second_upper_estimate": 0.002244953811255441
     },
     {
       "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": 28859.35586800948,
-      "required_pages_per_second_upper_estimate": 0.3701768303127138
+      "required_token_positions_per_second": 28875.637578935904,
+      "required_pages_per_second_upper_estimate": 0.37038567461853883
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06710471910156812,
-    "producer_p95_seconds": 0.1270958289999271,
-    "gpu_update_seconds": 0.9580255403637659,
-    "producer_batches_per_second": 14.902081603030386,
-    "gpu_batches_per_second": 1.0438135079575188,
-    "producer_headroom": 14.276574780288119,
+    "producer_batch_seconds": 0.04377005024999647,
+    "producer_p95_seconds": 0.12176520599996365,
+    "gpu_update_seconds": 0.9574853516020219,
+    "producer_batches_per_second": 22.84667242300186,
+    "gpu_batches_per_second": 1.0444024008585036,
+    "producer_headroom": 21.87535417787415,
     "sustainable": true,
-    "buffer_seconds": 60.35560904291725,
+    "buffer_seconds": 60.321577150927375,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 66,
+    "upload_pages_per_commit": 70,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -14,3 +14,4 @@
 {"step": 926400, "loss": 0.6199615597724915, "lr": 3e-05, "grad_norm": 0.3679225866314959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135598825732, "tokens_per_second": 152803.25678106054, "data_wait_s": 9.728999884828227e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 926500, "loss": 0.5330778956413269, "lr": 3e-05, "grad_norm": 0.352464082133268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135613285555, "tokens_per_second": 151386.98464105203, "data_wait_s": 6.169000016598147e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 926600, "loss": 0.5047388076782227, "lr": 3e-05, "grad_norm": 0.31476835165476447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135627853641, "tokens_per_second": 152753.400048464, "data_wait_s": 7.586000037917984e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 926700, "loss": 0.5236449241638184, "lr": 3e-05, "grad_norm": 0.334416242186587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135642531350, "tokens_per_second": 154168.46086430972, "data_wait_s": 8.987999990495155e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d05129bf7758828732a2f1247bc1c608ce8891fea39392c2ffdb445b60a707f2
-size 34053
+oid sha256:e89347087321dcb7867ff009bad16424c95bd0433bada8bdab954feaf76b4cfc
+size 36171