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": 76815.96162074867,
-      "required_pages_per_second_upper_estimate": 0.0053143064625107
+      "required_token_positions_per_second": 76886.9352816517,
+      "required_pages_per_second_upper_estimate": 0.005319216585053554
     },
     {
       "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": 9601.995202593584,
-      "required_pages_per_second_upper_estimate": 0.004020154878865072
+      "required_token_positions_per_second": 9610.866910206463,
+      "required_pages_per_second_upper_estimate": 0.004023869277580262
     },
     {
       "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": 19203.990405187167,
-      "required_pages_per_second_upper_estimate": 0.0039996847581166665
+      "required_token_positions_per_second": 19221.733820412926,
+      "required_pages_per_second_upper_estimate": 0.004003380243582866
     },
     {
       "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": 28805.985607780753,
-      "required_pages_per_second_upper_estimate": 0.0017268748529932004
+      "required_token_positions_per_second": 28832.60073061939,
+      "required_pages_per_second_upper_estimate": 0.0017284703889684373
     },
     {
       "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": 28805.985607780753,
-      "required_pages_per_second_upper_estimate": 0.0022395386768647714
+      "required_token_positions_per_second": 28832.60073061939,
+      "required_pages_per_second_upper_estimate": 0.002241607885597922
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28805.985607780753,
-      "required_pages_per_second_upper_estimate": 0.36949225391902046
+      "required_token_positions_per_second": 28832.60073061939,
+      "required_pages_per_second_upper_estimate": 0.36983364413770203
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.30087703720303693,
-    "producer_p95_seconds": 0.28703162399688154,
-    "gpu_update_seconds": 0.9598005211990396,
-    "producer_batches_per_second": 3.323616881155284,
-    "gpu_batches_per_second": 1.0418831600036442,
-    "producer_headroom": 3.190009214798768,
+    "producer_batch_seconds": 0.12112048089849736,
+    "producer_p95_seconds": 0.26220909600306186,
+    "gpu_update_seconds": 0.9589145376899221,
+    "producer_batches_per_second": 8.256241987992356,
+    "gpu_batches_per_second": 1.0428458018887221,
+    "producer_headroom": 7.917030468971813,
     "sustainable": true,
-    "buffer_seconds": 60.46743283553949,
+    "buffer_seconds": 60.41161587446509,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 452113.07448166056,
-    "upload_bytes_per_second": 4945804.51397623,
+    "generation_bytes_per_second": 450976.5159175186,
+    "upload_bytes_per_second": 4950490.843476476,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.01294113004077,
+    "estimated_full_shards_per_hour": 96.76906211990517,
     "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": 1.24713432200042,
-      "bytes": 5975341,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.473726465999789,
       "bytes": 9735362,
@@ -562,6 +556,12 @@
       "bytes": 11661576,
       "pages": 28,
       "shards": 3
+    },
+    {
+      "seconds": 1.2733937859993603,
+      "bytes": 6517566,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -232,3 +232,4 @@
 {"step": 908200, "loss": 0.5213196873664856, "lr": 3e-05, "grad_norm": 0.3281043705806001, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132934596573, "tokens_per_second": 154829.57237913038, "data_wait_s": 1.3740998838329688e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 908300, "loss": 0.5920027494430542, "lr": 3e-05, "grad_norm": 0.32380916206789356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132949354105, "tokens_per_second": 153057.10831939772, "data_wait_s": 1.5220000932458788e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 908400, "loss": 0.5087893605232239, "lr": 3e-05, "grad_norm": 0.3203193908856255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132964035956, "tokens_per_second": 152927.8710040776, "data_wait_s": 1.669399716774933e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 908500, "loss": 0.526980996131897, "lr": 3e-05, "grad_norm": 0.3245610635813911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132978672808, "tokens_per_second": 151831.38066959972, "data_wait_s": 1.5254001482389867e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:41ece52a23ee28517cbf42972f6509855b5cadf3edfe3915cacba2c6b498e4cc
-size 498909
+oid sha256:43baf985c2adbbfc788586b92417c000a906baf5e0f47307e526ead3f55c968a
+size 501027