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": 76058.0489074925,
-      "required_pages_per_second_upper_estimate": 0.005261872302407856
+      "required_token_positions_per_second": 76191.38572053821,
+      "required_pages_per_second_upper_estimate": 0.0052710968525183895
     },
     {
       "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": 9507.256113436562,
-      "required_pages_per_second_upper_estimate": 0.0039804896006121766
+      "required_token_positions_per_second": 9523.923215067276,
+      "required_pages_per_second_upper_estimate": 0.003987467768016297
     },
     {
       "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": 19014.512226873125,
-      "required_pages_per_second_upper_estimate": 0.003960221450449439
+      "required_token_positions_per_second": 19047.846430134552,
+      "required_pages_per_second_upper_estimate": 0.003967164085906739
     },
     {
       "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": 28521.76834030969,
-      "required_pages_per_second_upper_estimate": 0.0017098364617829497
+      "required_token_positions_per_second": 28571.769645201828,
+      "required_pages_per_second_upper_estimate": 0.0017128339636636614
     },
     {
       "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": 28521.76834030969,
-      "required_pages_per_second_upper_estimate": 0.0022174420344585367
+      "required_token_positions_per_second": 28571.769645201828,
+      "required_pages_per_second_upper_estimate": 0.0022213294159814036
     },
     {
       "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": 28521.76834030969,
-      "required_pages_per_second_upper_estimate": 0.36584661998062734
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28571.769645201828,
+      "required_pages_per_second_upper_estimate": 0.36648798303256536
     }
   ],
   "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": 0.06730667486716158,
-    "producer_p95_seconds": 0.16763282399915624,
-    "gpu_update_seconds": 0.9693648609060892,
-    "producer_batches_per_second": 14.857367444961875,
-    "gpu_batches_per_second": 1.031603310919766,
-    "producer_headroom": 14.402209926716125,
+    "producer_batch_seconds": 0.06730369210930576,
+    "producer_p95_seconds": 0.1659525160002886,
+    "gpu_update_seconds": 0.9676684483784868,
+    "producer_batches_per_second": 14.858025892189277,
+    "gpu_batches_per_second": 1.0334118071904597,
+    "producer_headroom": 14.377642861062178,
     "sustainable": true,
-    "buffer_seconds": 60.10062137617753,
+    "buffer_seconds": 60.96311224784466,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 159,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 665024.8487394642,
-    "upload_bytes_per_second": 5285813.480805551,
+    "generation_bytes_per_second": 647166.7064032437,
+    "upload_bytes_per_second": 5308166.279126671,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.34942577666256,
+    "estimated_full_shards_per_hour": 69.43345496212474,
     "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": [
@@ -382,6 +382,12 @@
       "bytes": 6213114,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.0006439850003517,
+      "bytes": 6195776,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -38,3 +38,4 @@
 {"step": 608200, "loss": 0.5756344199180603, "lr": 0.0001, "grad_norm": 0.1575766573149425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89022439208, "tokens_per_second": 153002.66190362268, "data_wait_s": 1.3965999642095994e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608300, "loss": 0.5493766069412231, "lr": 0.0001, "grad_norm": 0.17303658078893253, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89037184969, "tokens_per_second": 152626.1485758194, "data_wait_s": 1.1683000593620818e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 608400, "loss": 0.5173752307891846, "lr": 0.0001, "grad_norm": 0.14122168409750036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89051901676, "tokens_per_second": 152737.989892279, "data_wait_s": 1.485699976910837e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 608500, "loss": 0.5443984866142273, "lr": 0.0001, "grad_norm": 0.1289834412303883, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89066479770, "tokens_per_second": 150885.8250828299, "data_wait_s": 1.8201999409939162e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57a0eda588633efca390b88941bfdeeef21d150ee6e5cb2fb82358a5eb9e5d4b
-size 85668
+oid sha256:95aecbe0180fd9b15205be762923cb7b5ee64b251d1f711bd6d0994713e380da
+size 87786