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": 76835.06580124181,
-      "required_pages_per_second_upper_estimate": 0.005315628134045022
+      "required_token_positions_per_second": 76381.24150641607,
+      "required_pages_per_second_upper_estimate": 0.0052842315163115355
     },
     {
       "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": 9604.383225155227,
-      "required_pages_per_second_upper_estimate": 0.0040211546940440495
+      "required_token_positions_per_second": 9547.65518830201,
+      "required_pages_per_second_upper_estimate": 0.003997403849629724
     },
     {
       "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": 19208.766450310453,
-      "required_pages_per_second_upper_estimate": 0.004000679482363067
+      "required_token_positions_per_second": 19095.31037660402,
+      "required_pages_per_second_upper_estimate": 0.003977049574247886
     },
     {
       "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": 28813.14967546568,
-      "required_pages_per_second_upper_estimate": 0.0017273043279120138
+      "required_token_positions_per_second": 28642.96556490603,
+      "required_pages_per_second_upper_estimate": 0.0017171020503400534
     },
     {
       "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": 28813.14967546568,
-      "required_pages_per_second_upper_estimate": 0.0022400956516158707
+      "required_token_positions_per_second": 28642.96556490603,
+      "required_pages_per_second_upper_estimate": 0.0022268645855806498
     },
     {
       "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": 28813.14967546568,
-      "required_pages_per_second_upper_estimate": 0.3695841468871061
+      "required_token_positions_per_second": 28642.96556490603,
+      "required_pages_per_second_upper_estimate": 0.36740120784630814
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.07043241782807286,
-    "producer_p95_seconds": 0.14841911900020932,
-    "gpu_update_seconds": 0.9595618775250454,
-    "producer_batches_per_second": 14.198007548754365,
-    "gpu_batches_per_second": 1.0421422770350723,
-    "producer_headroom": 13.623866780597506,
+    "producer_batch_seconds": 0.0756846081953313,
+    "producer_p95_seconds": 0.1523534629995993,
+    "gpu_update_seconds": 0.9652631790988472,
+    "producer_batches_per_second": 13.212726125490947,
+    "gpu_batches_per_second": 1.035986891091798,
+    "producer_headroom": 12.753758024453784,
     "sustainable": true,
-    "buffer_seconds": 60.45239828407786,
+    "buffer_seconds": 60.81158028322737,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 475119.54357190005,
-    "upload_bytes_per_second": 2126612.9639993,
+    "generation_bytes_per_second": 472661.9534740035,
+    "upload_bytes_per_second": 2247675.591218118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.9495938336158,
+    "estimated_full_shards_per_hour": 101.42225220837668,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 11612692,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.2162191410006926,
+      "bytes": 6401722,
+      "pages": 45,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 527800, "loss": 0.6265311241149902, "lr": 0.0003, "grad_norm": 0.17604901438591505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77252273821, "tokens_per_second": 152148.98688793337, "data_wait_s": 1.5140999494178686e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 527900, "loss": 0.6602481603622437, "lr": 0.0003, "grad_norm": 0.19039419559360507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77266993109, "tokens_per_second": 153603.68486456753, "data_wait_s": 1.0407999980088789e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 528000, "loss": 0.5436892509460449, "lr": 0.0003, "grad_norm": 0.1983220608434485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77281614079, "tokens_per_second": 152006.00289369648, "data_wait_s": 1.8050000107905362e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 528100, "loss": 0.57123202085495, "lr": 0.0003, "grad_norm": 0.166388769357232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77296205947, "tokens_per_second": 152000.73141821977, "data_wait_s": 8.165000508597586e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:15746e6b3dabd73c4f8b8c65da73b4066a0f448dba0730bacd38e038f3cfd625
-size 63705
+oid sha256:32630f01aaf98b94f0b09d71daeb2fdaa73527b4e0c9695d3b687458ca31f354
+size 65823