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": 75505.95576166017,
-      "required_pages_per_second_upper_estimate": 0.006544664877325398
+      "required_token_positions_per_second": 74880.50437144592,
+      "required_pages_per_second_upper_estimate": 0.006490452336013682
     },
     {
       "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": 9438.24447020752,
-      "required_pages_per_second_upper_estimate": 0.0037650568334958996
+      "required_token_positions_per_second": 9360.06304643074,
+      "required_pages_per_second_upper_estimate": 0.003733869094634889
     },
     {
       "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": 18876.48894041504,
-      "required_pages_per_second_upper_estimate": 0.004366743501506215
+      "required_token_positions_per_second": 18720.12609286148,
+      "required_pages_per_second_upper_estimate": 0.004330571708616827
     },
     {
       "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": 28314.73341062256,
-      "required_pages_per_second_upper_estimate": 0.0017242187087817152
+      "required_token_positions_per_second": 28080.18913929222,
+      "required_pages_per_second_upper_estimate": 0.0017099361932163865
     },
     {
       "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": 28314.73341062256,
-      "required_pages_per_second_upper_estimate": 0.002179746558148361
+      "required_token_positions_per_second": 28080.18913929222,
+      "required_pages_per_second_upper_estimate": 0.002161690690881254
     },
     {
       "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": 28314.73341062256,
-      "required_pages_per_second_upper_estimate": 0.3515742256431523
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28080.18913929222,
+      "required_pages_per_second_upper_estimate": 0.34866197076241007
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11149532289041986,
-    "producer_p95_seconds": 0.3313197570023476,
-    "gpu_update_seconds": 0.9764527745695664,
-    "producer_batches_per_second": 8.968986089065124,
-    "gpu_batches_per_second": 1.0241150683818925,
-    "producer_headroom": 8.757791351743485,
+    "producer_batch_seconds": 0.08966958124193525,
+    "producer_p95_seconds": 0.3211770429988974,
+    "gpu_update_seconds": 0.9846087525568885,
+    "producer_batches_per_second": 11.152053864307954,
+    "gpu_batches_per_second": 1.0156318409755576,
+    "producer_headroom": 10.980409843783482,
     "sustainable": true,
-    "buffer_seconds": 60.540072023313115,
+    "buffer_seconds": 60.0611339059702,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300080.8519505204,
+    "generation_bytes_per_second": 305497.33358106535,
     "upload_bytes_per_second": 5175765.956355495,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.39036530386647,
+    "estimated_full_shards_per_hour": 65.55261617254229,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -325,3 +325,4 @@
 {"step": 387500, "loss": 0.5626535154879093, "lr": 0.0003, "grad_norm": 0.16091944570950384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56719988737, "tokens_per_second": 151672.9857508115, "data_wait_s": 1.1283998901490122e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 387600, "loss": 0.5485879797488451, "lr": 0.0003, "grad_norm": 0.15187291385905843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56734533574, "tokens_per_second": 148368.783878829, "data_wait_s": 1.2472002708818763e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 387700, "loss": 0.5797650143504143, "lr": 0.0003, "grad_norm": 0.1589285284853014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56748958865, "tokens_per_second": 146837.07462577018, "data_wait_s": 1.0078998457174748e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 387800, "loss": 0.5995310693979263, "lr": 0.0003, "grad_norm": 0.13778129882127946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56763541221, "tokens_per_second": 149192.33030781275, "data_wait_s": 1.1115000233985484e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ccf01716e477edc4b3d5b846c6cb46122895d3245c4a010f16b48b04785b9abb
-size 697449
+oid sha256:44d2363134c5810b0e71cbfb730ca6d5c277ed86b3443c63840a1f8b22da527d
+size 699567