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": 76488.66292141876,
-      "required_pages_per_second_upper_estimate": 0.005291663179053444
+      "required_token_positions_per_second": 76539.1181573115,
+      "required_pages_per_second_upper_estimate": 0.005295153789344782
     },
     {
       "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": 9561.082865177345,
-      "required_pages_per_second_upper_estimate": 0.004003025737535648
+      "required_token_positions_per_second": 9567.389769663938,
+      "required_pages_per_second_upper_estimate": 0.004005666306741043
     },
     {
       "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": 19122.16573035469,
-      "required_pages_per_second_upper_estimate": 0.003982642836210846
+      "required_token_positions_per_second": 19134.779539327876,
+      "required_pages_per_second_upper_estimate": 0.003985269959971449
     },
     {
       "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": 28683.248595532034,
-      "required_pages_per_second_upper_estimate": 0.001719516956517456
+      "required_token_positions_per_second": 28702.169308991815,
+      "required_pages_per_second_upper_estimate": 0.0017206512243991133
     },
     {
       "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": 28683.248595532034,
-      "required_pages_per_second_upper_estimate": 0.002229996414025496
+      "required_token_positions_per_second": 28702.169308991815,
+      "required_pages_per_second_upper_estimate": 0.0022314674162735707
     },
     {
       "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": 28683.248595532034,
-      "required_pages_per_second_upper_estimate": 0.3679179153106301
+      "required_token_positions_per_second": 28702.169308991815,
+      "required_pages_per_second_upper_estimate": 0.3681606099074129
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4664612544999187,
-    "producer_p95_seconds": 0.18640244100060954,
-    "gpu_update_seconds": 0.963907554191986,
-    "producer_batches_per_second": 2.1438007773487526,
-    "gpu_batches_per_second": 1.0374438872805278,
-    "producer_headroom": 2.0664257639691144,
+    "producer_batch_seconds": 0.05324185890626154,
+    "producer_p95_seconds": 0.1821749840000848,
+    "gpu_update_seconds": 0.9632721381564158,
+    "producer_batches_per_second": 18.78221423035991,
+    "gpu_batches_per_second": 1.038128230215271,
+    "producer_headroom": 18.09238366099065,
     "sustainable": true,
-    "buffer_seconds": 60.72617591409512,
+    "buffer_seconds": 60.6861447038542,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 305761.15100217436,
+    "generation_bytes_per_second": 304487.3509285632,
     "upload_bytes_per_second": 3214491.3619761616,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.60922524975703,
+    "estimated_full_shards_per_hour": 65.33589740650818,
     "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": [

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -87,3 +87,4 @@
 {"step": 653700, "loss": 0.5879846215248108, "lr": 0.0001, "grad_norm": 0.19696340840845797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95684610244, "tokens_per_second": 153054.70798256705, "data_wait_s": 1.1722000635927543e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 653800, "loss": 0.5489779710769653, "lr": 0.0001, "grad_norm": 0.18465427895519032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95699267514, "tokens_per_second": 132744.69051038052, "data_wait_s": 9.800000043469481e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 653900, "loss": 0.5059468746185303, "lr": 0.0001, "grad_norm": 0.17095308878730586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95713960871, "tokens_per_second": 152575.27706675188, "data_wait_s": 8.784998499322683e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 654000, "loss": 0.5774896740913391, "lr": 0.0001, "grad_norm": 0.19368160866251724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95728641982, "tokens_per_second": 153102.53756921992, "data_wait_s": 1.619199974811636e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a4e16d9ed0e0a934f4ae792f342d4be496088cb22d86b90f2130d5f72f65454e
-size 189450
+oid sha256:7b4efa8355291ff639d5cba3b3ad484106a0f211c8c5f52163acfa1fcc9da55f
+size 191568