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": 74297.30676053936,
-      "required_pages_per_second_upper_estimate": 0.006439902245201134
+      "required_token_positions_per_second": 76476.08831885221,
+      "required_pages_per_second_upper_estimate": 0.006628753508604317
     },
     {
       "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": 9287.16334506742,
-      "required_pages_per_second_upper_estimate": 0.0037047883138133955
+      "required_token_positions_per_second": 9559.511039856527,
+      "required_pages_per_second_upper_estimate": 0.003813431881225677
     },
     {
       "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": 18574.32669013484,
-      "required_pages_per_second_upper_estimate": 0.0042968435827778675
+      "required_token_positions_per_second": 19119.022079713053,
+      "required_pages_per_second_upper_estimate": 0.004422849274845343
     },
     {
       "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": 27861.49003520226,
-      "required_pages_per_second_upper_estimate": 0.0016966185652028294
+      "required_token_positions_per_second": 28678.53311956958,
+      "required_pages_per_second_upper_estimate": 0.0017463722023470286
     },
     {
       "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": 27861.49003520226,
-      "required_pages_per_second_upper_estimate": 0.0021448546284505415
+      "required_token_positions_per_second": 28678.53311956958,
+      "required_pages_per_second_upper_estimate": 0.002207752866805152
     },
     {
       "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": 27861.49003520226,
-      "required_pages_per_second_upper_estimate": 0.3459464598284299
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28678.53311956958,
+      "required_pages_per_second_upper_estimate": 0.35609140046897175
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10504269463280025,
-    "producer_p95_seconds": 0.33738547699613264,
-    "gpu_update_seconds": 0.9923374509068783,
-    "producer_batches_per_second": 9.51993856874787,
-    "gpu_batches_per_second": 1.0077217171297113,
-    "producer_headroom": 9.446991572101338,
+    "producer_batch_seconds": 0.11805967920315652,
+    "producer_p95_seconds": 0.3413104540013592,
+    "gpu_update_seconds": 0.9640660449656553,
+    "producer_batches_per_second": 8.470292370346058,
+    "gpu_batches_per_second": 1.0372733333177655,
+    "producer_headroom": 8.165921265182291,
     "sustainable": true,
-    "buffer_seconds": 60.532584505319576,
+    "buffer_seconds": 60.73616083283628,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373092.5294340365,
+    "generation_bytes_per_second": 372935.095843462,
     "upload_bytes_per_second": 4272189.235532882,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.05697166696379,
+    "estimated_full_shards_per_hour": 80.0231900832929,
     "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-v1/1790563279934888633-82973089/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 189900, "loss": 0.6188155319541693, "lr": 0.0003, "grad_norm": 0.14061967457514632, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27793615160, "tokens_per_second": 151529.41461533174, "data_wait_s": 1.0442003258503973e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 190000, "loss": 0.5794018357992172, "lr": 0.0003, "grad_norm": 0.143421867137207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27808256418, "tokens_per_second": 150552.32028538815, "data_wait_s": 7.763999747112393e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 190100, "loss": 0.5956941954791546, "lr": 0.0003, "grad_norm": 0.15725801850204055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27822802377, "tokens_per_second": 120622.25517734191, "data_wait_s": 8.716000593267381e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 190200, "loss": 0.6160158794373274, "lr": 0.0003, "grad_norm": 0.1624706316048518, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27837334859, "tokens_per_second": 149570.92182152078, "data_wait_s": 1.1028001608792692e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:328d49eb24371811eb4e79eeaf351eb65a1d525e879155802b20a24d5ff0774b
-size 749064
+oid sha256:b4d9b4ea384fe59ca973799ffd6e935e70ccd4289fcaba96bf25ed6a3a1c5572
+size 751182