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": 75488.05887415804,
-      "required_pages_per_second_upper_estimate": 0.00654311362047596
+      "required_token_positions_per_second": 74433.3051105829,
+      "required_pages_per_second_upper_estimate": 0.006451690237498241
     },
     {
       "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": 9436.007359269755,
-      "required_pages_per_second_upper_estimate": 0.0037641644164950355
+      "required_token_positions_per_second": 9304.163138822863,
+      "required_pages_per_second_upper_estimate": 0.0037115697857120085
     },
     {
       "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": 18872.01471853951,
-      "required_pages_per_second_upper_estimate": 0.004365708469018926
+      "required_token_positions_per_second": 18608.326277645727,
+      "required_pages_per_second_upper_estimate": 0.004304708788976207
     },
     {
       "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": 28308.022077809266,
-      "required_pages_per_second_upper_estimate": 0.0017238100238250276
+      "required_token_positions_per_second": 27912.48941646859,
+      "required_pages_per_second_upper_estimate": 0.0016997241599488746
     },
     {
       "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": 28308.022077809266,
-      "required_pages_per_second_upper_estimate": 0.0021792299011702356
+      "required_token_positions_per_second": 27912.48941646859,
+      "required_pages_per_second_upper_estimate": 0.0021487807019957466
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28308.022077809266,
-      "required_pages_per_second_upper_estimate": 0.35149089335099726
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27912.48941646859,
+      "required_pages_per_second_upper_estimate": 0.34657970145980843
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11303850597596465,
-    "producer_p95_seconds": 0.38194473299517995,
-    "gpu_update_seconds": 0.9766842743023484,
-    "producer_batches_per_second": 8.846542966629706,
-    "gpu_batches_per_second": 1.0238723263096523,
-    "producer_headroom": 8.640279397447278,
+    "producer_batch_seconds": 0.5988106238672231,
+    "producer_p95_seconds": 0.4064376739988802,
+    "gpu_update_seconds": 0.9905243343751153,
+    "producer_batches_per_second": 1.6699770514120578,
+    "gpu_batches_per_second": 1.0095663128063002,
+    "producer_headroom": 1.6541529072716463,
     "sustainable": true,
-    "buffer_seconds": 60.5544250067456,
+    "buffer_seconds": 60.42198439688204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 423831.4820798095,
-    "upload_bytes_per_second": 5556732.519288,
+    "generation_bytes_per_second": 381831.0718069277,
+    "upload_bytes_per_second": 5449450.029230074,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.94436976237976,
+    "estimated_full_shards_per_hour": 81.93205943733093,
     "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": 1.0771120079989487,
-      "bytes": 10854426,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.2815551900021092,
       "bytes": 7356201,
@@ -562,6 +556,12 @@
       "bytes": 7610947,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.7598877909986186,
+      "bytes": 961682,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 75600, "loss": 0.6527904272079468, "lr": 0.0003, "grad_norm": 0.1283468365167278, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11063941581, "tokens_per_second": 151668.4325913146, "data_wait_s": 9.27999644773081e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 75700, "loss": 0.6094121038913727, "lr": 0.0003, "grad_norm": 0.1716657095901446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11078685161, "tokens_per_second": 150022.6171831484, "data_wait_s": 1.651100319577381e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 75800, "loss": 0.5683665107935667, "lr": 0.0003, "grad_norm": 0.14402322371072332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11093339500, "tokens_per_second": 148663.41233548458, "data_wait_s": 1.1731004633475095e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 75900, "loss": 0.5978897251188755, "lr": 0.0003, "grad_norm": 0.13849490037262083, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11107875624, "tokens_per_second": 148662.9505346653, "data_wait_s": 1.391399564454332e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03be31adc9de4b5a5f78fd2382990de99d127dc37a65a1fa74d1ce39f0321d9d
-size 763890
+oid sha256:3d83fb670566027d3a64fbed152df2bb0e058d9b542ea00903b07855c6c64377
+size 766008