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": 76603.37079774599,
-      "required_pages_per_second_upper_estimate": 0.0052995989360967045
+      "required_token_positions_per_second": 77327.56690843176,
+      "required_pages_per_second_upper_estimate": 0.005349700503400433
     },
     {
       "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": 9575.421349718248,
-      "required_pages_per_second_upper_estimate": 0.004009028961591319
+      "required_token_positions_per_second": 9665.94586355397,
+      "required_pages_per_second_upper_estimate": 0.004046929685167526
     },
     {
       "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": 19150.842699436496,
-      "required_pages_per_second_upper_estimate": 0.003988615492608056
+      "required_token_positions_per_second": 19331.89172710794,
+      "required_pages_per_second_upper_estimate": 0.004026323230488081
     },
     {
       "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": 28726.264049154743,
-      "required_pages_per_second_upper_estimate": 0.001722095667281343
+      "required_token_positions_per_second": 28997.837590661908,
+      "required_pages_per_second_upper_estimate": 0.0017383760864259096
     },
     {
       "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": 28726.264049154743,
-      "required_pages_per_second_upper_estimate": 0.002233340676339677
+      "required_token_positions_per_second": 28997.837590661908,
+      "required_pages_per_second_upper_estimate": 0.002254454324666091
     },
     {
       "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": 28726.264049154743,
-      "required_pages_per_second_upper_estimate": 0.36846967136330655
+      "required_token_positions_per_second": 28997.837590661908,
+      "required_pages_per_second_upper_estimate": 0.37195312516080997
     }
   ],
   "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": 0.13780444442977569,
-    "producer_p95_seconds": 0.3173430500028189,
-    "gpu_update_seconds": 0.9624641739938866,
-    "producer_batches_per_second": 7.2566600020625165,
-    "gpu_batches_per_second": 1.038999712426025,
-    "producer_headroom": 6.984275274839575,
+    "producer_batch_seconds": 0.1331795151408528,
+    "producer_p95_seconds": 0.3114297770007397,
+    "gpu_update_seconds": 0.9534504051744674,
+    "producer_batches_per_second": 7.508662266433272,
+    "gpu_batches_per_second": 1.048822250819658,
+    "producer_headroom": 7.159137080249038,
     "sustainable": true,
-    "buffer_seconds": 60.63524296161486,
+    "buffer_seconds": 60.06737552599145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 450225.8879258704,
+    "generation_bytes_per_second": 450349.58844739257,
     "upload_bytes_per_second": 5079199.718115679,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.60799482662281,
+    "estimated_full_shards_per_hour": 96.63453807894071,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -348,3 +348,4 @@
 {"step": 639200, "loss": 0.575522243976593, "lr": 0.0001, "grad_norm": 0.2166116215513019, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93562733938, "tokens_per_second": 150187.81810608323, "data_wait_s": 1.72159998328425e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639300, "loss": 0.5429550409317017, "lr": 0.0001, "grad_norm": 0.1663941566307602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93577285796, "tokens_per_second": 151397.99076137244, "data_wait_s": 1.624900323804468e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 639400, "loss": 0.5889376401901245, "lr": 0.0001, "grad_norm": 0.18591583123044766, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93592017464, "tokens_per_second": 152868.0829272865, "data_wait_s": 1.2397998943924904e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 639500, "loss": 0.5777553915977478, "lr": 0.0001, "grad_norm": 0.17600662552134663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93606628231, "tokens_per_second": 142253.00815013997, "data_wait_s": 1.0637006198521703e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05fe333866522386fe210a24a89804bd6b1243b432b8d6a03734a38ef71ab78d
-size 746946
+oid sha256:eff42e733c0afc7653ea3f3ae0b9d63cac0efd440d0e2b89d7b04ad71206f28f
+size 749064