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": 75839.6536586954,
-      "required_pages_per_second_upper_estimate": 0.006573588965290824
+      "required_token_positions_per_second": 75200.1559666526,
+      "required_pages_per_second_upper_estimate": 0.006518158926136637
     },
     {
       "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": 9479.956707336925,
-      "required_pages_per_second_upper_estimate": 0.0037816964685403404
+      "required_token_positions_per_second": 9400.019495831575,
+      "required_pages_per_second_upper_estimate": 0.003749808319703038
     },
     {
       "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": 18959.91341467385,
-      "required_pages_per_second_upper_estimate": 0.004386042285405389
+      "required_token_positions_per_second": 18800.03899166315,
+      "required_pages_per_second_upper_estimate": 0.004349058151335609
     },
     {
       "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": 28439.870122010776,
-      "required_pages_per_second_upper_estimate": 0.0017318388779636732
+      "required_token_positions_per_second": 28200.058487494724,
+      "required_pages_per_second_upper_estimate": 0.0017172356076160535
     },
     {
       "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": 28439.870122010776,
-      "required_pages_per_second_upper_estimate": 0.0021893799285916137
+      "required_token_positions_per_second": 28200.058487494724,
+      "required_pages_per_second_upper_estimate": 0.002170918565125476
     },
     {
       "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": 28439.870122010776,
-      "required_pages_per_second_upper_estimate": 0.35312800479296197
+      "required_token_positions_per_second": 28200.058487494724,
+      "required_pages_per_second_upper_estimate": 0.3501503468901837
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.05270264038279393,
-    "producer_p95_seconds": 0.15259611099963877,
-    "gpu_update_seconds": 0.9721563383161193,
-    "producer_batches_per_second": 18.974381411191583,
-    "gpu_batches_per_second": 1.0286411357787462,
-    "producer_headroom": 18.44606515451745,
+    "producer_batch_seconds": 0.05523596989062085,
+    "producer_p95_seconds": 0.13455849199999648,
+    "gpu_update_seconds": 0.9804234984924044,
+    "producer_batches_per_second": 18.104144853077006,
+    "gpu_batches_per_second": 1.0199673932108915,
+    "producer_headroom": 17.749729034067016,
     "sustainable": true,
-    "buffer_seconds": 60.273692975599396,
+    "buffer_seconds": 60.786256906529076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 66,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -16,3 +16,4 @@
 {"step": 196600, "loss": 0.5984502658247948, "lr": 0.0003, "grad_norm": 0.14993225711578184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28774382558, "tokens_per_second": 150908.8984723203, "data_wait_s": 2.100000028804061e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196700, "loss": 0.6545633338391781, "lr": 0.0003, "grad_norm": 0.1603858787251228, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28789175919, "tokens_per_second": 151396.29797880666, "data_wait_s": 1.4754999938304536e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196800, "loss": 0.6597261168062687, "lr": 0.0003, "grad_norm": 0.1597154006396669, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28803846116, "tokens_per_second": 150012.79519459582, "data_wait_s": 1.6419000075984513e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 196900, "loss": 0.5936473049223423, "lr": 0.0003, "grad_norm": 0.14001345759618117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28818393057, "tokens_per_second": 149003.66544643656, "data_wait_s": 2.1523000214074273e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a327cfb08f08e3d83acb2ef2aa918468a046587e6bb5f0846ec2d42316e170c3
-size 38289
+oid sha256:c7714017f5743113ee9bf4d7324ec9fbc02af51abc13de81fe2a13c6317f8730
+size 40407