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": 76039.93691425954,
-      "required_pages_per_second_upper_estimate": 0.006590949010269853
+      "required_token_positions_per_second": 76396.46607357146,
+      "required_pages_per_second_upper_estimate": 0.0066218520541841165
     },
     {
       "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": 9504.992114282442,
-      "required_pages_per_second_upper_estimate": 0.0037916834666835977
+      "required_token_positions_per_second": 9549.558259196432,
+      "required_pages_per_second_upper_estimate": 0.0038094615682130336
     },
     {
       "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": 19009.984228564885,
-      "required_pages_per_second_upper_estimate": 0.004397625286982853
+      "required_token_positions_per_second": 19099.116518392864,
+      "required_pages_per_second_upper_estimate": 0.004418244473559832
     },
     {
       "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": 28514.976342847327,
-      "required_pages_per_second_upper_estimate": 0.0017364124527607322
+      "required_token_positions_per_second": 28648.674777589298,
+      "required_pages_per_second_upper_estimate": 0.001744553985975043
     },
     {
       "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": 28514.976342847327,
-      "required_pages_per_second_upper_estimate": 0.0021951618133789373
+      "required_token_positions_per_second": 28648.674777589298,
+      "required_pages_per_second_upper_estimate": 0.002205454288986329
     },
     {
       "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": 28514.976342847327,
-      "required_pages_per_second_upper_estimate": 0.35406057269140057
+      "required_token_positions_per_second": 28648.674777589298,
+      "required_pages_per_second_upper_estimate": 0.355720659791019
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.16087034039105674,
-    "producer_p95_seconds": 0.40553524200367974,
-    "gpu_update_seconds": 0.9695957544406381,
-    "producer_batches_per_second": 6.21618626260825,
-    "gpu_batches_per_second": 1.0313576512893274,
-    "producer_headroom": 6.027187809037176,
+    "producer_batch_seconds": 0.12558740502356613,
+    "producer_p95_seconds": 0.3785384349976084,
+    "gpu_update_seconds": 0.9650708179223674,
+    "producer_batches_per_second": 7.9625819150602934,
+    "gpu_batches_per_second": 1.036193387499613,
+    "producer_headroom": 7.684455441541088,
     "sustainable": true,
-    "buffer_seconds": 60.114936775319556,
+    "buffer_seconds": 60.799461529109145,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 496828.5312707336,
-    "upload_bytes_per_second": 4578151.474632579,
+    "generation_bytes_per_second": 516869.3861208552,
+    "upload_bytes_per_second": 4479140.786293985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.60783723441607,
+    "estimated_full_shards_per_hour": 110.90813815802805,
     "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": 1.3426188599987654,
-      "bytes": 12956908,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 1.2346246180022717,
       "bytes": 6315095,
@@ -562,6 +556,12 @@
       "bytes": 5813402,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.100297655997565,
+      "bytes": 2702936,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -418,3 +418,4 @@
 {"step": 196800, "loss": 0.6592029295861721, "lr": 0.0003, "grad_norm": 0.15857606671365335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28803846116, "tokens_per_second": 150832.67363795283, "data_wait_s": 1.2368996976874769e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196900, "loss": 0.5936013609170914, "lr": 0.0003, "grad_norm": 0.13963440142091707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28818393057, "tokens_per_second": 148291.17903467067, "data_wait_s": 1.1382995580788702e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197000, "loss": 0.6028390862047672, "lr": 0.0003, "grad_norm": 0.1438688130975058, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28832906369, "tokens_per_second": 149346.31997957057, "data_wait_s": 2.1495005057659e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 197100, "loss": 0.6232964936643839, "lr": 0.0003, "grad_norm": 0.15152785592092718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28847548768, "tokens_per_second": 147647.33111312086, "data_wait_s": 9.79300239123404e-06, "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 sha256:44f51029d0e6d2a2ab5ea08efb0d91ce02c24e5dc7a0bb8ba12b80d42851783c
-size 895989
+oid sha256:0fb35826e1f9f6da51079f674e888244082eb5bcfeff4cc1e4e8ec0fba3b9ac0
+size 898107