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": 75875.88202138119,
-      "required_pages_per_second_upper_estimate": 0.005249269574541265
+      "required_token_positions_per_second": 76532.62015115091,
+      "required_pages_per_second_upper_estimate": 0.00529470424220114
     },
     {
       "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": 9484.485252672648,
-      "required_pages_per_second_upper_estimate": 0.003970955916719971
+      "required_token_positions_per_second": 9566.577518893864,
+      "required_pages_per_second_upper_estimate": 0.004005326234305337
     },
     {
       "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": 18968.970505345296,
-      "required_pages_per_second_upper_estimate": 0.003950736310871153
+      "required_token_positions_per_second": 19133.15503778773,
+      "required_pages_per_second_upper_estimate": 0.003984931619141623
     },
     {
       "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": 28453.455758017946,
-      "required_pages_per_second_upper_estimate": 0.0017057412267818333
+      "required_token_positions_per_second": 28699.73255668159,
+      "required_pages_per_second_upper_estimate": 0.0017205051448188206
     },
     {
       "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": 28453.455758017946,
-      "required_pages_per_second_upper_estimate": 0.002212131031660648
+      "required_token_positions_per_second": 28699.73255668159,
+      "required_pages_per_second_upper_estimate": 0.002231277969499591
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28453.455758017946,
-      "required_pages_per_second_upper_estimate": 0.36497037952332506
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28699.73255668159,
+      "required_pages_per_second_upper_estimate": 0.3681293538651581
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10825819981252494,
-    "producer_p95_seconds": 0.2082491770006527,
-    "gpu_update_seconds": 0.9716921640426409,
-    "producer_batches_per_second": 9.237175583297525,
-    "gpu_batches_per_second": 1.0291325143959038,
-    "producer_headroom": 8.975691132176216,
+    "producer_batch_seconds": 0.43225089722661636,
+    "producer_p95_seconds": 0.20604718300000968,
+    "gpu_update_seconds": 0.963353924828239,
+    "producer_batches_per_second": 2.313471195586043,
+    "gpu_batches_per_second": 1.0380400953660875,
+    "producer_headroom": 2.2286915562448932,
     "sustainable": true,
-    "buffer_seconds": 60.24491417064374,
+    "buffer_seconds": 60.691297264179056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 368103.7932970754,
-    "upload_bytes_per_second": 3881562.6979871034,
+    "generation_bytes_per_second": 334843.1839406554,
+    "upload_bytes_per_second": 4025596.560948777,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.98650502380558,
+    "estimated_full_shards_per_hour": 71.84955252327677,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 1096275,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.9842258310000034,
+      "bytes": 13303551,
+      "pages": 43,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 806700, "loss": 0.536081075668335, "lr": 3e-05, "grad_norm": 0.2757851306726722, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118077248182, "tokens_per_second": 151394.7045601352, "data_wait_s": 1.7426999875169713e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806800, "loss": 0.5422860980033875, "lr": 3e-05, "grad_norm": 0.26394925850910905, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118091949073, "tokens_per_second": 151874.19530871272, "data_wait_s": 1.65940000442788e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 806900, "loss": 0.5263350605964661, "lr": 3e-05, "grad_norm": 0.2738297735929711, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118106612072, "tokens_per_second": 151285.86171342045, "data_wait_s": 1.7207999917445704e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 807000, "loss": 0.5280994772911072, "lr": 3e-05, "grad_norm": 0.3087932084461166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118121195350, "tokens_per_second": 151071.38485212842, "data_wait_s": 2.2965999960433692e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c0b12f87b8678d518e9ab9cd37fb43e6420f0640ca30a41fb3c8d8ddd74303e
-size 147090
+oid sha256:3311f296f53b6466e6540c286a11f6a1b57927bf8cbf443712672711599518b5
+size 149208