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": 74374.89355670854,
-      "required_pages_per_second_upper_estimate": 0.006446627272051141
+      "required_token_positions_per_second": 75070.22431610992,
+      "required_pages_per_second_upper_estimate": 0.006506896779976354
     },
     {
       "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": 9296.861694588568,
-      "required_pages_per_second_upper_estimate": 0.0037086571304406285
+      "required_token_positions_per_second": 9383.77803951374,
+      "required_pages_per_second_upper_estimate": 0.0037433293599464417
     },
     {
       "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": 18593.723389177136,
-      "required_pages_per_second_upper_estimate": 0.004301330667731059
+      "required_token_positions_per_second": 18767.55607902748,
+      "required_pages_per_second_upper_estimate": 0.004341543801177084
     },
     {
       "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": 27890.585083765705,
-      "required_pages_per_second_upper_estimate": 0.001698390301010421
+      "required_token_positions_per_second": 28151.33411854122,
+      "required_pages_per_second_upper_estimate": 0.0017142685491837923
     },
     {
       "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": 27890.585083765705,
-      "required_pages_per_second_upper_estimate": 0.0021470944458292047
+      "required_token_positions_per_second": 28151.33411854122,
+      "required_pages_per_second_upper_estimate": 0.0021671676283257497
     },
     {
       "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": 27890.585083765705,
-      "required_pages_per_second_upper_estimate": 0.3463077229567243
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28151.33411854122,
+      "required_pages_per_second_upper_estimate": 0.3495453532977541
     }
   ],
   "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": 0.09173785664077627,
-    "producer_p95_seconds": 0.24914866499966593,
-    "gpu_update_seconds": 0.9913022590584911,
-    "producer_batches_per_second": 10.900625288377547,
-    "gpu_batches_per_second": 1.0087740554024054,
-    "producer_headroom": 10.80581447351878,
+    "producer_batch_seconds": 0.08562565134394617,
+    "producer_p95_seconds": 0.2541814050018729,
+    "gpu_update_seconds": 0.9821204168718345,
+    "producer_batches_per_second": 11.678743277328671,
+    "gpu_batches_per_second": 1.0182050824125153,
+    "producer_headroom": 11.469932216069168,
     "sustainable": true,
-    "buffer_seconds": 60.46943780256796,
+    "buffer_seconds": 60.89146584605374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 371376.3090516956,
+    "generation_bytes_per_second": 383232.709090746,
     "upload_bytes_per_second": 5862472.530889289,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.68871072447921,
+    "estimated_full_shards_per_hour": 82.2328181699923,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -187,3 +187,4 @@
 {"step": 133700, "loss": 0.5746112670749426, "lr": 0.0003, "grad_norm": 0.12353973311740744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19566523091, "tokens_per_second": 149479.2568496617, "data_wait_s": 1.4605000615119934e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133800, "loss": 0.5904359146952629, "lr": 0.0003, "grad_norm": 0.12785646706601875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19581087503, "tokens_per_second": 147423.86574178244, "data_wait_s": 1.207899913424626e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 133900, "loss": 0.6582215577363968, "lr": 0.0003, "grad_norm": 0.14935984275103895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19595797030, "tokens_per_second": 150520.7337200414, "data_wait_s": 1.3793000107398257e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 134000, "loss": 0.6526389606297016, "lr": 0.0003, "grad_norm": 0.1691735713082882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19610482155, "tokens_per_second": 150119.77806600573, "data_wait_s": 1.133299883804284e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d6f3d1bf6042a76ec9508759e8247bec29e9683b0ffbc39d505d3c7a2af5a8eb
-size 402816
+oid sha256:a3e5c73010b0ece6c3f152e3817589322533a24f08c1e72876155964716f0cca
+size 404934