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": 74859.70170006962,
-      "required_pages_per_second_upper_estimate": 0.006488649213182674
+      "required_token_positions_per_second": 75620.33489322805,
+      "required_pages_per_second_upper_estimate": 0.006554578970558443
     },
     {
       "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": 9357.462712508703,
-      "required_pages_per_second_upper_estimate": 0.0037328317825549317
+      "required_token_positions_per_second": 9452.541861653506,
+      "required_pages_per_second_upper_estimate": 0.0037707602767087546
     },
     {
       "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": 18714.925425017405,
-      "required_pages_per_second_upper_estimate": 0.004329368625639728
+      "required_token_positions_per_second": 18905.083723307012,
+      "required_pages_per_second_upper_estimate": 0.004373358401277281
     },
     {
       "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": 28072.38813752611,
-      "required_pages_per_second_upper_estimate": 0.0017094611531375237
+      "required_token_positions_per_second": 28357.625584960515,
+      "required_pages_per_second_upper_estimate": 0.0017268306171610499
     },
     {
       "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": 28072.38813752611,
-      "required_pages_per_second_upper_estimate": 0.002161090148170744
+      "required_token_positions_per_second": 28357.625584960515,
+      "required_pages_per_second_upper_estimate": 0.002183048516462035
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28072.38813752611,
-      "required_pages_per_second_upper_estimate": 0.34856510842874844
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28357.625584960515,
+      "required_pages_per_second_upper_estimate": 0.3521068028975566
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07387341263282288,
-    "producer_p95_seconds": 0.12809065600004033,
-    "gpu_update_seconds": 0.9848823642845403,
-    "producer_batches_per_second": 13.536669883796428,
-    "gpu_batches_per_second": 1.0153496866871423,
-    "producer_headroom": 13.33202743969276,
+    "producer_batch_seconds": 0.11425693936718773,
+    "producer_p95_seconds": 0.1362277019998146,
+    "gpu_update_seconds": 0.9749758461675696,
+    "producer_batches_per_second": 8.752203634531975,
+    "gpu_batches_per_second": 1.0256664346412223,
+    "producer_headroom": 8.533187144408691,
     "sustainable": true,
-    "buffer_seconds": 60.07782422135696,
+    "buffer_seconds": 60.448502462389314,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 119,
+    "upload_pages_per_commit": 123,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 839586.303108899,
-    "upload_bytes_per_second": 3672933.616484457,
+    "generation_bytes_per_second": 843447.3124252327,
+    "upload_bytes_per_second": 3562201.029186526,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.07783803916087,
+    "estimated_full_shards_per_hour": 90.49207939895504,
     "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": [
@@ -214,6 +214,12 @@
       "bytes": 5863999,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.9417934939999668,
+      "bytes": 2584042,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -1,3 +1,4 @@
 {"step": 100, "loss": 2.8035809621214867, "lr": 2.9999999999999997e-05, "grad_norm": 1.586405728010338, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14758037, "tokens_per_second": 46373.26368537362, "data_wait_s": 8.257999979832675e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 200, "loss": 2.3905331194400787, "lr": 5.9999999999999995e-05, "grad_norm": 1.9926730765732787, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29475318, "tokens_per_second": 151060.5326281325, "data_wait_s": 9.137000006376184e-06, "gpu_peak_gb": 85.43514251708984}
 {"step": 300, "loss": 2.0513596944510937, "lr": 8.999999999999999e-05, "grad_norm": 4.616072880620013, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44215422, "tokens_per_second": 150363.0206983736, "data_wait_s": 1.6987000208246172e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 400, "loss": 1.7191495224833488, "lr": 0.00011999999999999999, "grad_norm": 2.6239859511966785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58978491, "tokens_per_second": 150971.93971749238, "data_wait_s": 1.4216999943528208e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae0a2047c4c103d5143a3c9eaf0ef2b1783b54e9e0e8ac06fb2ca28b7faeef29
-size 6375
+oid sha256:300964d53c1ce329e942229bc9acfeaccdfd276ce7dcc4c72fc6d2cabaf0cbb6
+size 8458