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": 76064.36120189368,
-      "required_pages_per_second_upper_estimate": 0.00659306604561919
+      "required_token_positions_per_second": 75730.98518207653,
+      "required_pages_per_second_upper_estimate": 0.0065641698571552365
     },
     {
       "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": 9508.04515023671,
-      "required_pages_per_second_upper_estimate": 0.0037929013683727104
+      "required_token_positions_per_second": 9466.373147759567,
+      "required_pages_per_second_upper_estimate": 0.003776277783532618
     },
     {
       "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": 19016.09030047342,
-      "required_pages_per_second_upper_estimate": 0.004399037819255698
+      "required_token_positions_per_second": 18932.746295519133,
+      "required_pages_per_second_upper_estimate": 0.0043797576505139124
     },
     {
       "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": 28524.135450710128,
-      "required_pages_per_second_upper_estimate": 0.0017369701943754517
+      "required_token_positions_per_second": 28399.1194432787,
+      "required_pages_per_second_upper_estimate": 0.0017293573754311776
     },
     {
       "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": 28524.135450710128,
-      "required_pages_per_second_upper_estimate": 0.0021958669068562493
+      "required_token_positions_per_second": 28399.1194432787,
+      "required_pages_per_second_upper_estimate": 0.0021862428285377203
     },
     {
       "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": 28524.135450710128,
-      "required_pages_per_second_upper_estimate": 0.3541742981574944
+      "required_token_positions_per_second": 28399.1194432787,
+      "required_pages_per_second_upper_estimate": 0.352622017746858
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08912981562468758,
-    "producer_p95_seconds": 0.3475798739964375,
-    "gpu_update_seconds": 0.9692844169729843,
-    "producer_batches_per_second": 11.219590133686033,
-    "gpu_batches_per_second": 1.0316889268920042,
-    "producer_headroom": 10.874973881405714,
+    "producer_batch_seconds": 0.1132474231878291,
+    "producer_p95_seconds": 0.3509319080039859,
+    "gpu_update_seconds": 0.9735513122236447,
+    "producer_batches_per_second": 8.83022299184174,
+    "gpu_batches_per_second": 1.027167225234328,
+    "producer_headroom": 8.596675180934925,
     "sustainable": true,
-    "buffer_seconds": 60.09563385232502,
+    "buffer_seconds": 60.36018135786597,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 381688.24503665016,
-    "upload_bytes_per_second": 4628569.859332334,
+    "generation_bytes_per_second": 423658.38289445604,
+    "upload_bytes_per_second": 4615418.862234146,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.90141213726643,
+    "estimated_full_shards_per_hour": 90.90722670674573,
     "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": [
-    {
-      "seconds": 0.8633549010010029,
-      "bytes": 2675452,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1101803599995037,
       "bytes": 11781038,
@@ -562,6 +556,12 @@
       "bytes": 13649971,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 0.8202600639997399,
+      "bytes": 1256174,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -335,3 +335,4 @@
 {"step": 188500, "loss": 0.6497993152588606, "lr": 0.0003, "grad_norm": 0.16283055332802274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27588149914, "tokens_per_second": 147988.9902466689, "data_wait_s": 1.0895993909798563e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 188600, "loss": 0.6287831570953131, "lr": 0.0003, "grad_norm": 0.14656467815856106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27602632371, "tokens_per_second": 148410.82226704736, "data_wait_s": 7.88000033935532e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 188700, "loss": 0.608249181881547, "lr": 0.0003, "grad_norm": 0.1537021722364676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27617295906, "tokens_per_second": 150804.7138387289, "data_wait_s": 6.722999387420714e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 188800, "loss": 0.5681229531764984, "lr": 0.0003, "grad_norm": 0.1350764465890865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27631983050, "tokens_per_second": 150737.48718055035, "data_wait_s": 1.2507000064942986e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:374da6adf8d79b6b8f0e5d628accda4aa4bfa6068bc7dbaabc3f2a76c8aa232f
-size 718629
+oid sha256:272fe674b3dae23e8e4a22d87ab459b7e21faa4968e6e32482ff8b81f1fc592e
+size 720747