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": 77084.74304473559,
-      "required_pages_per_second_upper_estimate": 0.005332901385080942
+      "required_token_positions_per_second": 77269.4542283766,
+      "required_pages_per_second_upper_estimate": 0.005345680133354222
     },
     {
       "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": 9635.592880591948,
-      "required_pages_per_second_upper_estimate": 0.004034221525043688
+      "required_token_positions_per_second": 9658.681778547076,
+      "required_pages_per_second_upper_estimate": 0.004043888364466484
     },
     {
       "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": 19271.185761183897,
-      "required_pages_per_second_upper_estimate": 0.004013679778710082
+      "required_token_positions_per_second": 19317.36355709415,
+      "required_pages_per_second_upper_estimate": 0.0040232973958078165
     },
     {
       "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": 28906.778641775847,
-      "required_pages_per_second_upper_estimate": 0.0017329172414791546
+      "required_token_positions_per_second": 28976.04533564123,
+      "required_pages_per_second_upper_estimate": 0.0017370696740122667
     },
     {
       "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": 28906.778641775847,
-      "required_pages_per_second_upper_estimate": 0.002247374891916187
+      "required_token_positions_per_second": 28976.04533564123,
+      "required_pages_per_second_upper_estimate": 0.002252760072692233
     },
     {
       "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": 28906.778641775847,
-      "required_pages_per_second_upper_estimate": 0.3707851187359814
+      "required_token_positions_per_second": 28976.04533564123,
+      "required_pages_per_second_upper_estimate": 0.37167359751210516
     }
   ],
   "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": 0.05218987320311541,
-    "producer_p95_seconds": 0.13707047500020053,
-    "gpu_update_seconds": 0.9564538595816875,
-    "producer_batches_per_second": 19.160805317693438,
-    "gpu_batches_per_second": 1.0455287413836751,
-    "producer_headroom": 18.32642619880121,
+    "producer_batch_seconds": 0.05988339016407096,
+    "producer_p95_seconds": 0.14353704700033632,
+    "gpu_update_seconds": 0.9541674745377451,
+    "producer_batches_per_second": 16.699121363372367,
+    "gpu_batches_per_second": 1.0480340471513754,
+    "producer_headroom": 15.933758458288317,
     "sustainable": true,
-    "buffer_seconds": 60.256593153646314,
+    "buffer_seconds": 60.11255089587794,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

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

@@ -32,3 +32,4 @@
 {"step": 928200, "loss": 0.5369636416435242, "lr": 3e-05, "grad_norm": 0.32165364746021524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135862068058, "tokens_per_second": 152026.8280368941, "data_wait_s": 6.794999990233919e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 928300, "loss": 0.5516103506088257, "lr": 3e-05, "grad_norm": 0.328012424760905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135876730454, "tokens_per_second": 153856.77058871454, "data_wait_s": 1.1735000043699984e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 928400, "loss": 0.584817111492157, "lr": 3e-05, "grad_norm": 0.34219711359203886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135891380783, "tokens_per_second": 153575.66740318137, "data_wait_s": 1.710800006549107e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 928500, "loss": 0.5752276182174683, "lr": 3e-05, "grad_norm": 0.35367433659115854, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135906124229, "tokens_per_second": 154919.63366021946, "data_wait_s": 1.2986999990971526e-05, "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:f71badfe28d365adb00fe7271fe68a46867b6e225cad875503b3fa0470413bc3
-size 72177
+oid sha256:41a67cd83f5685fca983ac2cd8e890d162ef2c5cadad91df2f53a3bd5732b299
+size 74295