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": 75328.32338872181,
-      "required_pages_per_second_upper_estimate": 0.005211388197822819
+      "required_token_positions_per_second": 76362.50154371379,
+      "required_pages_per_second_upper_estimate": 0.005282935042209088
     },
     {
       "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": 9416.040423590226,
-      "required_pages_per_second_upper_estimate": 0.003942299496073722
+      "required_token_positions_per_second": 9545.312692964224,
+      "required_pages_per_second_upper_estimate": 0.003996423095748659
     },
     {
       "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": 18832.080847180452,
-      "required_pages_per_second_upper_estimate": 0.003922225805098466
+      "required_token_positions_per_second": 19090.625385928448,
+      "required_pages_per_second_upper_estimate": 0.003976073814241678
     },
     {
       "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": 28248.121270770676,
-      "required_pages_per_second_upper_estimate": 0.0016934317377989676
+      "required_token_positions_per_second": 28635.938078892672,
+      "required_pages_per_second_upper_estimate": 0.0017166807632839026
     },
     {
       "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": 28248.121270770676,
-      "required_pages_per_second_upper_estimate": 0.0021961671784481323
+      "required_token_positions_per_second": 28635.938078892672,
+      "required_pages_per_second_upper_estimate": 0.0022263182294537556
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28248.121270770676,
-      "required_pages_per_second_upper_estimate": 0.3623365691919123
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28635.938078892672,
+      "required_pages_per_second_upper_estimate": 0.3673110668012554
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.17165305735949232,
-    "producer_p95_seconds": 0.2672137240006123,
-    "gpu_update_seconds": 0.9787553563290723,
-    "producer_batches_per_second": 5.82570456584239,
-    "gpu_batches_per_second": 1.0217057751291478,
-    "producer_headroom": 5.701939548208971,
+    "producer_batch_seconds": 0.26526038791416795,
+    "producer_p95_seconds": 0.29864754500158597,
+    "gpu_update_seconds": 0.9655000623282926,
+    "producer_batches_per_second": 3.769880636394065,
+    "gpu_batches_per_second": 1.0357327140803194,
+    "producer_headroom": 3.639819989408693,
     "sustainable": true,
-    "buffer_seconds": 60.682832092402485,
+    "buffer_seconds": 60.82650392668243,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421786.14276060986,
+    "generation_bytes_per_second": 430131.2574076824,
     "upload_bytes_per_second": 3893563.4988446785,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.50548755754205,
+    "estimated_full_shards_per_hour": 92.29615489647726,
     "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-v1/1791300814123050567-72be911d/train.jsonl

@@ -225,3 +225,4 @@
 {"step": 822500, "loss": 0.5585592985153198, "lr": 3e-05, "grad_norm": 0.2840605367832263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120390929173, "tokens_per_second": 150758.69757127998, "data_wait_s": 1.357400105916895e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822600, "loss": 0.47821998596191406, "lr": 3e-05, "grad_norm": 0.26762885928740976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120405455371, "tokens_per_second": 150056.61543589283, "data_wait_s": 1.017099930322729e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 822700, "loss": 0.5082652568817139, "lr": 3e-05, "grad_norm": 0.2780539188763553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120420075097, "tokens_per_second": 150939.55668971944, "data_wait_s": 1.1304000508971512e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 822800, "loss": 0.5378347039222717, "lr": 3e-05, "grad_norm": 0.2849654067059827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120434734215, "tokens_per_second": 151493.9209661294, "data_wait_s": 1.4625999028794467e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b966f0343111b6937a972a9b44352839be15a6f3bab4d3b519256f7d35607d5
-size 484083
+oid sha256:b106a0d82e7073746a77c9ed42147519687c1760e95317ab08d0995e5fc23d5e
+size 486201