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": 76729.15523615335,
-      "required_pages_per_second_upper_estimate": 0.005308300995405834
+      "required_token_positions_per_second": 76569.23155556619,
+      "required_pages_per_second_upper_estimate": 0.005297237104108751
     },
     {
       "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": 9591.14440451917,
-      "required_pages_per_second_upper_estimate": 0.004015611876301744
+      "required_token_positions_per_second": 9571.153944445774,
+      "required_pages_per_second_upper_estimate": 0.00400724228811729
     },
     {
       "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": 19182.28880903834,
-      "required_pages_per_second_upper_estimate": 0.003995164887948442
+      "required_token_positions_per_second": 19142.307888891548,
+      "required_pages_per_second_upper_estimate": 0.003986837916649633
     },
     {
       "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": 28773.433213557506,
-      "required_pages_per_second_upper_estimate": 0.0017249233866641449
+      "required_token_positions_per_second": 28713.461833337322,
+      "required_pages_per_second_upper_estimate": 0.0017213281939909401
     },
     {
       "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": 28773.433213557506,
-      "required_pages_per_second_upper_estimate": 0.002237007871396764
+      "required_token_positions_per_second": 28713.461833337322,
+      "required_pages_per_second_upper_estimate": 0.0022323453603708683
     },
     {
       "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": 28773.433213557506,
-      "required_pages_per_second_upper_estimate": 0.3690747067579624
+      "required_token_positions_per_second": 28713.461833337322,
+      "required_pages_per_second_upper_estimate": 0.3683054582847491
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08894894996872438,
-    "producer_p95_seconds": 0.1714517740001611,
-    "gpu_update_seconds": 0.9608863771936946,
-    "producer_batches_per_second": 11.24240365233781,
-    "gpu_batches_per_second": 1.0407057730598057,
-    "producer_headroom": 10.802672516444039,
+    "producer_batch_seconds": 0.06930161364063991,
+    "producer_p95_seconds": 0.1801968849995319,
+    "gpu_update_seconds": 0.9628932993338934,
+    "producer_batches_per_second": 14.429678437005096,
+    "gpu_batches_per_second": 1.0385366693192029,
+    "producer_headroom": 13.894240678534976,
     "sustainable": true,
-    "buffer_seconds": 60.53584176320276,
+    "buffer_seconds": 60.66227785803529,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379471.5890971472,
-    "upload_bytes_per_second": 4873139.889233944,
+    "generation_bytes_per_second": 384306.2724841303,
+    "upload_bytes_per_second": 4886517.818467076,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.42576937375844,
+    "estimated_full_shards_per_hour": 82.46317988293582,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -358,6 +358,12 @@
       "bytes": 5808038,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.466896263000308,
+      "bytes": 7685001,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -75,3 +75,4 @@
 {"step": 727500, "loss": 0.5392053127288818, "lr": 5e-05, "grad_norm": 0.2323101730866226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106485432700, "tokens_per_second": 150974.56391772357, "data_wait_s": 1.2364999747660477e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727600, "loss": 0.561463475227356, "lr": 5e-05, "grad_norm": 0.21850676530010643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106500106885, "tokens_per_second": 152043.26874381435, "data_wait_s": 1.2847999641962815e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727700, "loss": 0.5557624101638794, "lr": 5e-05, "grad_norm": 0.23337268929924065, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106514784425, "tokens_per_second": 151908.00363559255, "data_wait_s": 1.4756999917153735e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 727800, "loss": 0.5481175184249878, "lr": 5e-05, "grad_norm": 0.23142053883470848, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106529601973, "tokens_per_second": 138000.22715871155, "data_wait_s": 2.3218999558594078e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2153d05fcf9e8a7c9ec0bbbf7b12f2a710e00c4862bf94922ef78a89a983a3ae
-size 164034
+oid sha256:15db3eb4c450c4dcdf0c373e42dcabe6fd85f2db1ba8298a879bb34c619c12d6
+size 166152