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": 75960.20764125376,
-      "required_pages_per_second_upper_estimate": 0.006584038278957706
+      "required_token_positions_per_second": 75704.39948043282,
+      "required_pages_per_second_upper_estimate": 0.006561865475917605
     },
     {
       "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": 9495.02595515672,
-      "required_pages_per_second_upper_estimate": 0.0037877078168009892
+      "required_token_positions_per_second": 9463.049935054103,
+      "required_pages_per_second_upper_estimate": 0.0037749521042979507
     },
     {
       "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": 18990.05191031344,
-      "required_pages_per_second_upper_estimate": 0.004393014295952201
+      "required_token_positions_per_second": 18926.099870108206,
+      "required_pages_per_second_upper_estimate": 0.004378220116968186
     },
     {
       "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": 28485.077865470157,
-      "required_pages_per_second_upper_estimate": 0.0017345917923536503
+      "required_token_positions_per_second": 28389.149805162306,
+      "required_pages_per_second_upper_estimate": 0.0017287502767765377
     },
     {
       "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": 28485.077865470157,
-      "required_pages_per_second_upper_estimate": 0.00219286014582616
+      "required_token_positions_per_second": 28389.149805162306,
+      "required_pages_per_second_upper_estimate": 0.0021854753381977954
     },
     {
       "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": 28485.077865470157,
-      "required_pages_per_second_upper_estimate": 0.35368933366614297
+      "required_token_positions_per_second": 28389.149805162306,
+      "required_pages_per_second_upper_estimate": 0.3524982282076847
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.10125668532029408,
-    "producer_p95_seconds": 0.3108638250014337,
-    "gpu_update_seconds": 0.9706134605134827,
-    "producer_batches_per_second": 9.875891125971688,
-    "gpu_batches_per_second": 1.030276253814748,
-    "producer_headroom": 9.585672861433775,
+    "producer_batch_seconds": 0.08154706267188772,
+    "producer_p95_seconds": 0.3077418809989467,
+    "gpu_update_seconds": 0.9738932017954431,
+    "producer_batches_per_second": 12.262857388543768,
+    "gpu_batches_per_second": 1.0268066335779191,
+    "producer_headroom": 11.942713445289797,
     "sustainable": true,
-    "buffer_seconds": 60.17803455183593,
+    "buffer_seconds": 60.38137851131747,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 583507.6223998398,
+    "generation_bytes_per_second": 575821.0277707783,
     "upload_bytes_per_second": 4943480.428053393,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.60357620237539,
+    "estimated_full_shards_per_hour": 61.77889406605964,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -280,3 +280,4 @@
 {"step": 383000, "loss": 0.543660243973136, "lr": 0.0003, "grad_norm": 0.19995132168341503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56060283476, "tokens_per_second": 148668.3106738177, "data_wait_s": 1.178499951492995e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 383100, "loss": 0.5857820343226194, "lr": 0.0003, "grad_norm": 0.15258547212936122, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56074807961, "tokens_per_second": 148263.606208162, "data_wait_s": 1.1571999493753538e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 383200, "loss": 0.5977911725640297, "lr": 0.0003, "grad_norm": 0.16053210512025626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56089396953, "tokens_per_second": 148709.62632530794, "data_wait_s": 6.9789966801181436e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 383300, "loss": 0.59818534553051, "lr": 0.0003, "grad_norm": 0.18368050872323263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56104023859, "tokens_per_second": 149688.41897770754, "data_wait_s": 1.0154002666240558e-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:a9cabc97700cb411d716484c3efcdd8531e034778df9a254051f550e761267db
-size 601356
+oid sha256:eed5db5994428cb596017114599c51e604a7b4893b8b7e22f59e75da720fd605
+size 603474