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": 77245.88576972042,
-      "required_pages_per_second_upper_estimate": 0.005344049612698029
+      "required_token_positions_per_second": 75656.85773028113,
+      "required_pages_per_second_upper_estimate": 0.005234116965876591
     },
     {
       "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": 9655.735721215053,
-      "required_pages_per_second_upper_estimate": 0.004042654911782239
+      "required_token_positions_per_second": 9457.107216285141,
+      "required_pages_per_second_upper_estimate": 0.003959493304602934
     },
     {
       "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": 19311.471442430106,
-      "required_pages_per_second_upper_estimate": 0.004022070223708808
+      "required_token_positions_per_second": 18914.214432570283,
+      "required_pages_per_second_upper_estimate": 0.003939332064926864
     },
     {
       "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": 28967.20716364516,
-      "required_pages_per_second_upper_estimate": 0.0017365398390962095
+      "required_token_positions_per_second": 28371.321648855424,
+      "required_pages_per_second_upper_estimate": 0.0017008174123490632
     },
     {
       "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": 28967.20716364516,
-      "required_pages_per_second_upper_estimate": 0.0022520729436945486
+      "required_token_positions_per_second": 28371.321648855424,
+      "required_pages_per_second_upper_estimate": 0.002205745465943028
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28967.20716364516,
-      "required_pages_per_second_upper_estimate": 0.37156023093142926
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28371.321648855424,
+      "required_pages_per_second_upper_estimate": 0.3639168513597238
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10887002392183831,
-    "producer_p95_seconds": 0.32176802100002533,
-    "gpu_update_seconds": 0.9544586001614678,
-    "producer_batches_per_second": 9.185264813737303,
-    "gpu_batches_per_second": 1.0477143794721195,
-    "producer_headroom": 8.766954996232093,
+    "producer_batch_seconds": 0.08830511370319982,
+    "producer_p95_seconds": 0.32464936400356237,
+    "gpu_update_seconds": 0.9745051831632029,
+    "producer_batches_per_second": 11.324372486072276,
+    "gpu_batches_per_second": 1.0261618073226066,
+    "producer_headroom": 11.035659683748198,
     "sustainable": true,
-    "buffer_seconds": 60.13089181017247,
+    "buffer_seconds": 60.41932135611858,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 310975.0827132075,
-    "upload_bytes_per_second": 5495795.574231083,
+    "generation_bytes_per_second": 315026.9510195452,
+    "upload_bytes_per_second": 5539116.282184017,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.72801362082643,
+    "estimated_full_shards_per_hour": 67.59745023669974,
     "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": 0.9367298329998448,
-      "bytes": 2789154,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.218474695997429,
       "bytes": 10329696,
@@ -562,6 +556,12 @@
       "bytes": 7506775,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.0540632260017446,
+      "bytes": 6792818,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -332,3 +332,4 @@
 {"step": 473400, "loss": 0.6197728514671326, "lr": 0.0003, "grad_norm": 0.1419205834011256, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69291051761, "tokens_per_second": 151552.33377451796, "data_wait_s": 1.4473000192083418e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 473500, "loss": 0.6271270513534546, "lr": 0.0003, "grad_norm": 0.16629573102673484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69305679802, "tokens_per_second": 152218.25866785712, "data_wait_s": 1.1481002729851753e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 473600, "loss": 0.557610273361206, "lr": 0.0003, "grad_norm": 0.19698444025243064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69320354325, "tokens_per_second": 153501.88476630833, "data_wait_s": 1.2992000847589225e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 473700, "loss": 0.5600315928459167, "lr": 0.0003, "grad_norm": 0.17488606683979363, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69334970703, "tokens_per_second": 125777.29565164473, "data_wait_s": 9.38499579206109e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:03c970b5bd90eedc74e2e8d61e6199b5b4fbf53f43c04c9b19fa7f798dd41e44
-size 712275
+oid sha256:a40da1cace23857e839814db3f3e5a7ef037cfbef9c613e984138b58f353e66f
+size 714393