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": 75648.58959868352,
-      "required_pages_per_second_upper_estimate": 0.006557028016816427
+      "required_token_positions_per_second": 75722.42029985205,
+      "required_pages_per_second_upper_estimate": 0.006563427474871514
     },
     {
       "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": 9456.07369983544,
-      "required_pages_per_second_upper_estimate": 0.003772169179765215
+      "required_token_positions_per_second": 9465.302537481506,
+      "required_pages_per_second_upper_estimate": 0.0037758507010856495
     },
     {
       "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": 18912.14739967088,
-      "required_pages_per_second_upper_estimate": 0.004374992458487605
+      "required_token_positions_per_second": 18930.605074963012,
+      "required_pages_per_second_upper_estimate": 0.004379262316822449
     },
     {
       "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": 28368.22109950632,
-      "required_pages_per_second_upper_estimate": 0.0017274758284065207
+      "required_token_positions_per_second": 28395.907612444516,
+      "required_pages_per_second_upper_estimate": 0.0017291617917845497
     },
     {
       "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": 28368.22109950632,
-      "required_pages_per_second_upper_estimate": 0.0021838641885020323
+      "required_token_positions_per_second": 28395.907612444516,
+      "required_pages_per_second_upper_estimate": 0.0021859955729091876
     },
     {
       "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": 28368.22109950632,
-      "required_pages_per_second_upper_estimate": 0.35223836372730943
+      "required_token_positions_per_second": 28395.907612444516,
+      "required_pages_per_second_upper_estimate": 0.35258213755720375
     }
   ],
   "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": 0.040149508164105896,
-    "producer_p95_seconds": 0.13641897900015465,
-    "gpu_update_seconds": 0.9746116932401216,
-    "producer_batches_per_second": 24.90690535766042,
-    "gpu_batches_per_second": 1.0260496636106162,
-    "producer_headroom": 24.274561204000882,
+    "producer_batch_seconds": 0.039259478234370704,
+    "producer_p95_seconds": 0.13552765300119063,
+    "gpu_update_seconds": 0.9736614295745649,
+    "producer_batches_per_second": 25.47155604132621,
+    "gpu_batches_per_second": 1.027051056584365,
+    "producer_headroom": 24.800671668686324,
     "sustainable": true,
-    "buffer_seconds": 60.425924980887544,
+    "buffer_seconds": 60.36700863362302,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 86700, "loss": 0.6090225968509912, "lr": 0.0003, "grad_norm": 0.13672678645286673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12687942064, "tokens_per_second": 148855.0959608192, "data_wait_s": 8.787999831838533e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 86800, "loss": 0.6323533430695534, "lr": 0.0003, "grad_norm": 0.13809571203695895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12702535251, "tokens_per_second": 149576.5669256539, "data_wait_s": 8.88800059328787e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 86900, "loss": 0.5960618462413549, "lr": 0.0003, "grad_norm": 0.15141781592707554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12717146045, "tokens_per_second": 149597.13523622364, "data_wait_s": 7.530999937443994e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 87000, "loss": 0.6178943142294884, "lr": 0.0003, "grad_norm": 0.13060739307729685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12731783694, "tokens_per_second": 150387.5653255563, "data_wait_s": 8.342998626176268e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:068c449fd20813c1178bd5f36b279a0fe1c88f6f7b02e7f2be171525308a4e7c
-size 40407
+oid sha256:7c5333e0375714b9f1de27c12ae486315fbf90ee1f5e9250151ee0569010381b
+size 42525