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": 75778.55798979758,
-      "required_pages_per_second_upper_estimate": 0.006568293347556318
+      "required_token_positions_per_second": 75938.26074882413,
+      "required_pages_per_second_upper_estimate": 0.0065821359779457975
     },
     {
       "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": 9472.319748724698,
-      "required_pages_per_second_upper_estimate": 0.0037786499715672162
+      "required_token_positions_per_second": 9492.282593603017,
+      "required_pages_per_second_upper_estimate": 0.003786613448860307
     },
     {
       "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": 18944.639497449396,
-      "required_pages_per_second_upper_estimate": 0.004382508933467274
+      "required_token_positions_per_second": 18984.565187206033,
+      "required_pages_per_second_upper_estimate": 0.004391745039124331
     },
     {
       "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": 28416.959246174094,
-      "required_pages_per_second_upper_estimate": 0.0017304437258292948
+      "required_token_positions_per_second": 28476.84778080905,
+      "required_pages_per_second_upper_estimate": 0.001734090623377706
     },
     {
       "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": 28416.959246174094,
-      "required_pages_per_second_upper_estimate": 0.0021876161859483425
+      "required_token_positions_per_second": 28476.84778080905,
+      "required_pages_per_second_upper_estimate": 0.002192226571126616
     },
     {
       "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": 28416.959246174094,
-      "required_pages_per_second_upper_estimate": 0.3528435283928392
+      "required_token_positions_per_second": 28476.84778080905,
+      "required_pages_per_second_upper_estimate": 0.3535871435589735
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,17 +158,17 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.05075358793749629,
-    "producer_p95_seconds": 0.1430304430000433,
-    "gpu_update_seconds": 0.972940129184384,
-    "producer_batches_per_second": 19.70304052654392,
-    "gpu_batches_per_second": 1.0278124727348847,
-    "producer_headroom": 19.169878795220793,
+    "producer_batch_seconds": 0.05404563364841408,
+    "producer_p95_seconds": 0.15243659799989473,
+    "gpu_update_seconds": 0.9708939771989924,
+    "producer_batches_per_second": 18.502882332832897,
+    "gpu_batches_per_second": 1.0299785800350496,
+    "producer_headroom": 17.9643370177691,
     "sustainable": true,
-    "buffer_seconds": 60.32228800943181,
+    "buffer_seconds": 60.195426586337526,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790696019647648547-3fdea0a8/train.jsonl

@@ -32,3 +32,4 @@
 {"step": 278200, "loss": 0.6077223289757967, "lr": 0.0003, "grad_norm": 0.1378774455084411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40720756769, "tokens_per_second": 151397.99093341763, "data_wait_s": 1.0745000054157572e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278300, "loss": 0.6166594307869673, "lr": 0.0003, "grad_norm": 0.17162817212766507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40735477528, "tokens_per_second": 151484.95224747804, "data_wait_s": 1.0196999937761575e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 278400, "loss": 0.6390067264437675, "lr": 0.0003, "grad_norm": 0.164547309035477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40750257836, "tokens_per_second": 152601.1102714849, "data_wait_s": 1.526100004412001e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 278500, "loss": 0.6061176229268312, "lr": 0.0003, "grad_norm": 0.16668151308511772, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40765085288, "tokens_per_second": 152982.57103350223, "data_wait_s": 1.4629999895987567e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790696019.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.22407.0.1790696019647648547-3fdea0a8

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:afa6b76144a27e70133dcd01c528205d837093a562c3f499f546f59699708c86
-size 72177
+oid sha256:f348630c760aa27f75b2c4c9b55fd460fff5279314ba9693a4f4b27d2d52c4a5
+size 74295