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": 74992.82310360523,
-      "required_pages_per_second_upper_estimate": 0.006500187839048027
+      "required_token_positions_per_second": 72849.88269933774,
+      "required_pages_per_second_upper_estimate": 0.0063144431960389215
     },
     {
       "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": 9374.102887950654,
-      "required_pages_per_second_upper_estimate": 0.00373946979733152
+      "required_token_positions_per_second": 9106.235337417218,
+      "required_pages_per_second_upper_estimate": 0.0036326134264469514
     },
     {
       "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": 18748.205775901308,
-      "required_pages_per_second_upper_estimate": 0.004337067449102548
+      "required_token_positions_per_second": 18212.470674834436,
+      "required_pages_per_second_upper_estimate": 0.004213134562086478
     },
     {
       "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": 28122.30866385196,
-      "required_pages_per_second_upper_estimate": 0.0017125010512780078
+      "required_token_positions_per_second": 27318.706012251652,
+      "required_pages_per_second_upper_estimate": 0.0016635658659728187
     },
     {
       "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": 28122.30866385196,
-      "required_pages_per_second_upper_estimate": 0.002164933168476169
+      "required_token_positions_per_second": 27318.706012251652,
+      "required_pages_per_second_upper_estimate": 0.0021030696118414624
     },
     {
       "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": 28122.30866385196,
-      "required_pages_per_second_upper_estimate": 0.3491849542924613
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27318.706012251652,
+      "required_pages_per_second_upper_estimate": 0.33920689884465094
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09806024578159622,
-    "producer_p95_seconds": 0.39671837799687637,
-    "gpu_update_seconds": 0.9831340780189348,
-    "producer_batches_per_second": 10.197812498117134,
-    "gpu_batches_per_second": 1.0171552612793677,
-    "producer_headroom": 10.02581698814636,
+    "producer_batch_seconds": 0.22824643225010277,
+    "producer_p95_seconds": 0.40646344300330384,
+    "gpu_update_seconds": 1.0120537915522305,
+    "producer_batches_per_second": 4.381229490168952,
+    "gpu_batches_per_second": 0.9880897718551668,
+    "producer_headroom": 4.434039917185934,
     "sustainable": true,
-    "buffer_seconds": 60.95431283717396,
+    "buffer_seconds": 60.723227493133834,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 380111.8026197752,
+    "generation_bytes_per_second": 386072.05331668543,
     "upload_bytes_per_second": 4014829.5622885916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.56314429230635,
+    "estimated_full_shards_per_hour": 82.84207534432814,
     "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-v1/1790738827394728502-c744ae20/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 349900, "loss": 0.584590258076787, "lr": 0.0003, "grad_norm": 0.17714293403909717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51216225970, "tokens_per_second": 147863.93379601528, "data_wait_s": 9.56900476012379e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 350000, "loss": 0.6504786228761077, "lr": 0.0003, "grad_norm": 0.1511979477375556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51230919415, "tokens_per_second": 148850.98533615225, "data_wait_s": 1.4363999071065336e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350100, "loss": 0.693084504455328, "lr": 0.0003, "grad_norm": 0.18748733395407843, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51245680972, "tokens_per_second": 135294.73832244033, "data_wait_s": 1.1290001566521823e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 350200, "loss": 0.6044878903776407, "lr": 0.0003, "grad_norm": 0.16243606073841022, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51260237399, "tokens_per_second": 147832.4141225818, "data_wait_s": 6.558002496603876e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:72f30e1aa68e9e7d16e526cfa697f4a6eb8dfdb507f418b8fc791317e1170e8b
-size 749064
+oid sha256:a5aa91e9178dec884af8692c4a9c9498da2d529986e1351e64820a62a96ee1b4
+size 751182