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": 75662.22778680972,
-      "required_pages_per_second_upper_estimate": 0.006558210140397534
+      "required_token_positions_per_second": 76009.09606751193,
+      "required_pages_per_second_upper_estimate": 0.006588275803839181
     },
     {
       "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": 9457.778473351214,
-      "required_pages_per_second_upper_estimate": 0.0037728492394092923
+      "required_token_positions_per_second": 9501.137008438991,
+      "required_pages_per_second_upper_estimate": 0.003790145607323676
     },
     {
       "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": 18915.55694670243,
-      "required_pages_per_second_upper_estimate": 0.0043757811971873745
+      "required_token_positions_per_second": 19002.274016877982,
+      "required_pages_per_second_upper_estimate": 0.004395841665204179
     },
     {
       "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": 28373.335420053645,
-      "required_pages_per_second_upper_estimate": 0.0017277872636950865
+      "required_token_positions_per_second": 28503.411025316975,
+      "required_pages_per_second_upper_estimate": 0.0017357081856016654
     },
     {
       "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": 28373.335420053645,
-      "required_pages_per_second_upper_estimate": 0.002184257903055111
+      "required_token_positions_per_second": 28503.411025316975,
+      "required_pages_per_second_upper_estimate": 0.002194271483220603
     },
     {
       "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": 28373.335420053645,
-      "required_pages_per_second_upper_estimate": 0.3523018664719774
+      "required_token_positions_per_second": 28503.411025316975,
+      "required_pages_per_second_upper_estimate": 0.35391697015430146
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.04218157585156135,
-    "producer_p95_seconds": 0.1213712989999749,
-    "gpu_update_seconds": 0.9744360185605463,
-    "producer_batches_per_second": 23.707032746216974,
-    "gpu_batches_per_second": 1.026234643375783,
-    "producer_headroom": 23.10098660110816,
+    "producer_batch_seconds": 0.04974418565624106,
+    "producer_p95_seconds": 0.11810652299993762,
+    "gpu_update_seconds": 0.9699891699082194,
+    "producer_batches_per_second": 20.10285195762446,
+    "gpu_batches_per_second": 1.0309393455337448,
+    "producer_headroom": 19.49954868316397,
     "sustainable": true,
-    "buffer_seconds": 60.41503315075387,
+    "buffer_seconds": 60.1393285343096,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 40,
+    "upload_pages_per_commit": 45,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -1,3 +1,4 @@
 {"step": 355100, "loss": 0.6090846359729767, "lr": 0.0003, "grad_norm": 0.15904267394337226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51976574318, "tokens_per_second": 50459.87599969507, "data_wait_s": 9.298000009039242e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 355200, "loss": 0.5521676354110241, "lr": 0.0003, "grad_norm": 0.14981194429073466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51991218815, "tokens_per_second": 150210.32586298898, "data_wait_s": 1.2212999990879325e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 355300, "loss": 0.5499131996184587, "lr": 0.0003, "grad_norm": 0.14731746978116014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52005888775, "tokens_per_second": 151265.82736267187, "data_wait_s": 1.427199993031536e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 355400, "loss": 0.6366731058806181, "lr": 0.0003, "grad_norm": 0.1852487765654712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52020506515, "tokens_per_second": 150339.57901114027, "data_wait_s": 1.3126999988344323e-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 sha256:58563b5f6d16794c85639ebe8f59885e6c6d30362f5ddc32440d55b3ce8cfa01
-size 6519
+oid sha256:45264dd2fe75005af9566b16c366ed0da31496304e8c71d8f158812d1d94866c
+size 8637