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": 75113.63464842916,
-      "required_pages_per_second_upper_estimate": 0.006510659477559305
+      "required_token_positions_per_second": 74607.8936121174,
+      "required_pages_per_second_upper_estimate": 0.0064668231263207395
     },
     {
       "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": 9389.204331053645,
-      "required_pages_per_second_upper_estimate": 0.0037454939887719982
+      "required_token_positions_per_second": 9325.986701514676,
+      "required_pages_per_second_upper_estimate": 0.003720275531161112
     },
     {
       "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": 18778.40866210729,
-      "required_pages_per_second_upper_estimate": 0.004344054355273659
+      "required_token_positions_per_second": 18651.97340302935,
+      "required_pages_per_second_upper_estimate": 0.004314805783270538
     },
     {
       "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": 28167.612993160932,
-      "required_pages_per_second_upper_estimate": 0.0017152598472394785
+      "required_token_positions_per_second": 27977.96010454403,
+      "required_pages_per_second_upper_estimate": 0.0017037109813545154
     },
     {
       "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": 28167.612993160932,
-      "required_pages_per_second_upper_estimate": 0.002168420821156784
+      "required_token_positions_per_second": 27977.96010454403,
+      "required_pages_per_second_upper_estimate": 0.0021538208167982582
     },
     {
       "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": 28167.612993160932,
-      "required_pages_per_second_upper_estimate": 0.3497474824386423
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27977.96010454403,
+      "required_pages_per_second_upper_estimate": 0.3473926282894077
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.09125675247659615,
-    "producer_p95_seconds": 0.2999369619974459,
-    "gpu_update_seconds": 0.9815528212033056,
-    "producer_batches_per_second": 10.95809321350178,
-    "gpu_batches_per_second": 1.018793872727175,
-    "producer_headroom": 10.755947308721469,
+    "producer_batch_seconds": 0.10696145578899063,
+    "producer_p95_seconds": 0.3041218110010959,
+    "gpu_update_seconds": 0.9882064273695766,
+    "producer_batches_per_second": 9.349162206363017,
+    "gpu_batches_per_second": 1.0119343209108806,
+    "producer_headroom": 9.238902182848665,
     "sustainable": true,
-    "buffer_seconds": 60.856274914604946,
+    "buffer_seconds": 60.28059206954418,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392176.1092518871,
-    "upload_bytes_per_second": 5280668.181750713,
+    "generation_bytes_per_second": 420557.6915749436,
+    "upload_bytes_per_second": 5263741.251223332,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.15186365287265,
+    "estimated_full_shards_per_hour": 90.2418905299781,
     "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": 1.0223479320011393,
-      "bytes": 6489223,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.180141814002127,
       "bytes": 7534736,
@@ -562,6 +556,12 @@
       "bytes": 1068713,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.4721264299987524,
+      "bytes": 7499280,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -252,3 +252,4 @@
 {"step": 340200, "loss": 0.5384630095213652, "lr": 0.0003, "grad_norm": 0.14425019723219684, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49795998935, "tokens_per_second": 148280.82035737755, "data_wait_s": 1.2831998901674524e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340300, "loss": 0.5752277784049511, "lr": 0.0003, "grad_norm": 0.15664839531034666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49810604865, "tokens_per_second": 148973.96658369276, "data_wait_s": 1.9014998542843387e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 340400, "loss": 0.6086094714701176, "lr": 0.0003, "grad_norm": 0.1441742419979587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49825273922, "tokens_per_second": 148742.5304692349, "data_wait_s": 2.0366001990623772e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 340500, "loss": 0.6149112656712532, "lr": 0.0003, "grad_norm": 0.1481979501688307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49840024880, "tokens_per_second": 150326.8567130313, "data_wait_s": 1.1689000530168414e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7c84480bcf81227a360964528955fa617cebce3a2964032470a7b05d227f34e
-size 542052
+oid sha256:129ff2965dea6458e052650e8e60a921f6ba81f8fd32d2e0aeabc6792561dd5d
+size 544170