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": 75515.18870997605,
-      "required_pages_per_second_upper_estimate": 0.006545465165884725
+      "required_token_positions_per_second": 76417.91916703463,
+      "required_pages_per_second_upper_estimate": 0.006623711553953137
     },
     {
       "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": 9439.398588747006,
-      "required_pages_per_second_upper_estimate": 0.00376551722863691
+      "required_token_positions_per_second": 9552.23989587933,
+      "required_pages_per_second_upper_estimate": 0.0038105313131798824
     },
     {
       "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": 18878.797177494012,
-      "required_pages_per_second_upper_estimate": 0.00436727747153085
+      "required_token_positions_per_second": 19104.47979175866,
+      "required_pages_per_second_upper_estimate": 0.004419485172462621
     },
     {
       "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": 28318.19576624102,
-      "required_pages_per_second_upper_estimate": 0.0017244295480733035
+      "required_token_positions_per_second": 28656.71968763799,
+      "required_pages_per_second_upper_estimate": 0.0017450438787886273
     },
     {
       "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": 28318.19576624102,
-      "required_pages_per_second_upper_estimate": 0.002180013099868284
+      "required_token_positions_per_second": 28656.71968763799,
+      "required_pages_per_second_upper_estimate": 0.002206073608431614
     },
     {
       "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": 28318.19576624102,
-      "required_pages_per_second_upper_estimate": 0.35161721651217476
+      "required_token_positions_per_second": 28656.71968763799,
+      "required_pages_per_second_upper_estimate": 0.35582055064924184
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.08323907945302267,
+    "producer_batch_seconds": 0.08297729971855006,
     "producer_p95_seconds": 0.23059817899775226,
-    "gpu_update_seconds": 0.976333387487914,
-    "producer_batches_per_second": 12.013587927343265,
-    "gpu_batches_per_second": 1.0242402982581387,
-    "producer_headroom": 11.729266996986958,
+    "gpu_update_seconds": 0.9647998899164606,
+    "producer_batches_per_second": 12.051488821543854,
+    "gpu_batches_per_second": 1.036484363702184,
+    "producer_headroom": 11.627275088354965,
     "sustainable": true,
-    "buffer_seconds": 60.53267002425067,
+    "buffer_seconds": 60.782393064737015,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300315.55757197586,
-    "upload_bytes_per_second": 3094975.390778224,
+    "generation_bytes_per_second": 287819.325473002,
+    "upload_bytes_per_second": 3044929.711684992,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.44072766656357,
+    "estimated_full_shards_per_hour": 61.75932715551896,
     "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": 1.1156518739990133,
-      "bytes": 7370564,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.0604308100009803,
       "bytes": 5273416,
@@ -562,6 +556,12 @@
       "bytes": 12643645,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 2.9626744119996147,
+      "bytes": 6098299,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 295500, "loss": 0.5683505292981863, "lr": 0.0003, "grad_norm": 0.15291408250282582, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43253262817, "tokens_per_second": 149585.33872645616, "data_wait_s": 1.1811996955657378e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 295600, "loss": 0.5909181497991085, "lr": 0.0003, "grad_norm": 0.15410086744153798, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43268035802, "tokens_per_second": 151579.548667534, "data_wait_s": 1.327800055150874e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 295700, "loss": 0.6116938721388578, "lr": 0.0003, "grad_norm": 0.16125981915337717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43282650490, "tokens_per_second": 149593.41374083722, "data_wait_s": 8.348000847036019e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 295800, "loss": 0.5968709141016006, "lr": 0.0003, "grad_norm": 0.17676553626767794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43297298767, "tokens_per_second": 146914.03649349697, "data_wait_s": 9.439001587452367e-06, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f3379d2df0bd5298e9020a97027db7ffac77e5d365254261aa9cf507d178c331
-size 330804
+oid sha256:14e4c662e6c0a6d3332e85a415ab5abc1e5a774c1b0f008da8920570a5b21b21
+size 332922