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": 74665.65900367181,
-      "required_pages_per_second_upper_estimate": 0.006471830084055629
+      "required_token_positions_per_second": 75631.85527665944,
+      "required_pages_per_second_upper_estimate": 0.006555577528196131
     },
     {
       "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": 9333.207375458976,
-      "required_pages_per_second_upper_estimate": 0.0037231559659561897
+      "required_token_positions_per_second": 9453.98190958243,
+      "required_pages_per_second_upper_estimate": 0.0037713347333582373
     },
     {
       "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": 18666.414750917953,
-      "required_pages_per_second_upper_estimate": 0.004318146534945524
+      "required_token_positions_per_second": 18907.96381916486,
+      "required_pages_per_second_upper_estimate": 0.004374024660766039
     },
     {
       "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": 27999.622126376933,
-      "required_pages_per_second_upper_estimate": 0.0017050300848322941
+      "required_token_positions_per_second": 28361.945728747287,
+      "required_pages_per_second_upper_estimate": 0.0017270936912516213
     },
     {
       "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": 27999.622126376933,
-      "required_pages_per_second_upper_estimate": 0.0021554884192032694
+      "required_token_positions_per_second": 28361.945728747287,
+      "required_pages_per_second_upper_estimate": 0.002183381093089664
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27999.622126376933,
-      "required_pages_per_second_upper_estimate": 0.34766159810244895
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28361.945728747287,
+      "required_pages_per_second_upper_estimate": 0.3521604446247971
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07951490426565044,
-    "producer_p95_seconds": 0.1986074689993984,
-    "gpu_update_seconds": 0.9874418974374055,
-    "producer_batches_per_second": 12.576258617618546,
-    "gpu_batches_per_second": 1.0127178141774065,
-    "producer_headroom": 12.418324672044779,
+    "producer_batch_seconds": 0.41426244714827476,
+    "producer_p95_seconds": 0.2175090969994926,
+    "gpu_update_seconds": 0.9748273360517843,
+    "producer_batches_per_second": 2.4139286746454123,
+    "gpu_batches_per_second": 1.0258226898418434,
+    "producer_headroom": 2.3531636593236014,
     "sustainable": true,
-    "buffer_seconds": 60.23395574368173,
+    "buffer_seconds": 60.439294835210625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 436141.2042163931,
-    "upload_bytes_per_second": 5971646.863384404,
+    "generation_bytes_per_second": 363597.42662447016,
+    "upload_bytes_per_second": 6014956.314295295,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.58574957722516,
+    "estimated_full_shards_per_hour": 78.01954363871172,
     "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": [
@@ -550,6 +550,12 @@
       "bytes": 2018917,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.1833946610004205,
+      "bytes": 10020680,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -100,3 +100,4 @@
 {"step": 50000, "loss": 0.678381010890007, "lr": 0.0003, "grad_norm": 0.14200276719499727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7317670920, "tokens_per_second": 151211.93643269406, "data_wait_s": 8.727000022190623e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 50100, "loss": 0.661159873008728, "lr": 0.0003, "grad_norm": 0.16064657262487617, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7332387355, "tokens_per_second": 139554.97944974128, "data_wait_s": 1.0337000276194885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 50200, "loss": 0.6411357149481773, "lr": 0.0003, "grad_norm": 0.1354083447785202, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7346867699, "tokens_per_second": 147604.6036964836, "data_wait_s": 1.146300019172486e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 50300, "loss": 0.6121636796742678, "lr": 0.0003, "grad_norm": 0.12666534850508873, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 7361338036, "tokens_per_second": 148150.24210174807, "data_wait_s": 1.0711000868468545e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f4c9f043e51355e1c38f941c6003c9242109c140060bdf44f1488c2aa99cbd5
-size 217767
+oid sha256:24db4217089c85676ce672aeb852a3ef99ca4168c538ee1c93f70d3d5ddfc0c0
+size 219885