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": 77074.64357832419,
-      "required_pages_per_second_upper_estimate": 0.005332202680560612
+      "required_token_positions_per_second": 77449.28410684623,
+      "required_pages_per_second_upper_estimate": 0.005358121181609557
     },
     {
       "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": 9634.330447290524,
-      "required_pages_per_second_upper_estimate": 0.004033692970582987
+      "required_token_positions_per_second": 9681.160513355779,
+      "required_pages_per_second_upper_estimate": 0.0040532997413215265
     },
     {
       "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": 19268.660894581048,
-      "required_pages_per_second_upper_estimate": 0.004013153915581918
+      "required_token_positions_per_second": 19362.321026711557,
+      "required_pages_per_second_upper_estimate": 0.00403266085112092
     },
     {
       "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": 28902.99134187157,
-      "required_pages_per_second_upper_estimate": 0.0017326901986327652
+      "required_token_positions_per_second": 29043.481540067336,
+      "required_pages_per_second_upper_estimate": 0.0017411123714984897
     },
     {
       "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": 28902.99134187157,
-      "required_pages_per_second_upper_estimate": 0.002247080446007201
+      "required_token_positions_per_second": 29043.481540067336,
+      "required_pages_per_second_upper_estimate": 0.002258002940965847
     },
     {
       "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": 28902.99134187157,
-      "required_pages_per_second_upper_estimate": 0.3707365393192952
+      "required_token_positions_per_second": 29043.481540067336,
+      "required_pages_per_second_upper_estimate": 0.3725385967351283
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.05216600211713285,
-    "producer_p95_seconds": 0.17736114799845382,
-    "gpu_update_seconds": 0.9565791883951654,
-    "producer_batches_per_second": 19.16957327407635,
-    "gpu_batches_per_second": 1.0453917586035724,
-    "producer_headroom": 18.337214844397607,
+    "producer_batch_seconds": 0.10060939578123396,
+    "producer_p95_seconds": 0.17857649299912737,
+    "gpu_update_seconds": 0.9519519883268065,
+    "producer_batches_per_second": 9.939429535730536,
+    "gpu_batches_per_second": 1.0504731459804446,
+    "producer_headroom": 9.461859709372872,
     "sustainable": true,
-    "buffer_seconds": 60.26448886889542,
+    "buffer_seconds": 60.924927252915616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306831.70074398513,
-    "upload_bytes_per_second": 4503042.943106597,
+    "generation_bytes_per_second": 347363.8969514814,
+    "upload_bytes_per_second": 4460920.571874349,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.83894030322709,
+    "estimated_full_shards_per_hour": 74.5362060681184,
     "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": [
@@ -556,6 +556,12 @@
       "bytes": 6104360,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1603049650002504,
+      "bytes": 1160533,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -80,3 +80,4 @@
 {"step": 448200, "loss": 0.6464706659317017, "lr": 0.0003, "grad_norm": 0.15260309219318896, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65602686693, "tokens_per_second": 152285.74856316112, "data_wait_s": 1.0685000233934261e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 448300, "loss": 0.6037876009941101, "lr": 0.0003, "grad_norm": 0.14330422360713538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65617219715, "tokens_per_second": 151601.41238243, "data_wait_s": 9.137000233749859e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 448400, "loss": 0.5554867386817932, "lr": 0.0003, "grad_norm": 0.14706922969069713, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65631727393, "tokens_per_second": 151249.41875372521, "data_wait_s": 1.0630999895511195e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 448500, "loss": 0.5858342051506042, "lr": 0.0003, "grad_norm": 0.15514609296353427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65646337332, "tokens_per_second": 152351.41002405842, "data_wait_s": 9.543000487610698e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a20a1e7872a4c27434bfd9f514037e36bea88d8709050d6c11ce390963bcb2f
-size 174624
+oid sha256:dba429965e0b891c5a53fcd9febd48e267a68b917812d8af956439987c0b6c83
+size 176742