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": 75228.61760565774,
-      "required_pages_per_second_upper_estimate": 0.006520625908870237
+      "required_token_positions_per_second": 73974.3960680793,
+      "required_pages_per_second_upper_estimate": 0.006411913164788363
     },
     {
       "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": 9403.577200707217,
-      "required_pages_per_second_upper_estimate": 0.0037512275413703598
+      "required_token_positions_per_second": 9246.799508509912,
+      "required_pages_per_second_upper_estimate": 0.003688686575917469
     },
     {
       "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": 18807.154401414435,
-      "required_pages_per_second_upper_estimate": 0.004350704176154635
+      "required_token_positions_per_second": 18493.599017019824,
+      "required_pages_per_second_upper_estimate": 0.00427816865636123
     },
     {
       "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": 28210.73160212165,
-      "required_pages_per_second_upper_estimate": 0.0017178855442993286
+      "required_token_positions_per_second": 27740.398525529738,
+      "required_pages_per_second_upper_estimate": 0.0016892447275818241
     },
     {
       "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": 28210.73160212165,
-      "required_pages_per_second_upper_estimate": 0.002171740211034529
+      "required_token_positions_per_second": 27740.398525529738,
+      "required_pages_per_second_upper_estimate": 0.0021355326688331983
     },
     {
       "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": 28210.73160212165,
-      "required_pages_per_second_upper_estimate": 0.3502828712532333
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27740.398525529738,
+      "required_pages_per_second_upper_estimate": 0.34444290854550996
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11253154135938814,
-    "producer_p95_seconds": 0.21601555200140865,
-    "gpu_update_seconds": 0.9800525697079289,
-    "producer_batches_per_second": 8.886397430621999,
-    "gpu_batches_per_second": 1.0203534288961824,
-    "producer_headroom": 8.709136637327026,
+    "producer_batch_seconds": 0.06076831097648494,
+    "producer_p95_seconds": 0.20148954599972058,
+    "gpu_update_seconds": 0.9966691709405435,
+    "producer_batches_per_second": 16.45594527692176,
+    "gpu_batches_per_second": 1.0033419605588012,
+    "producer_headroom": 16.40113333619256,
     "sustainable": true,
-    "buffer_seconds": 60.76325932189159,
+    "buffer_seconds": 60.79681942737316,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353505.4464497467,
+    "generation_bytes_per_second": 386333.6953831713,
     "upload_bytes_per_second": 5276205.103530173,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.8540396224909,
+    "estimated_full_shards_per_hour": 82.89821764108042,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -87,3 +87,4 @@
 {"step": 203700, "loss": 0.6329334229230881, "lr": 0.0003, "grad_norm": 0.1702569485568475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29814096252, "tokens_per_second": 147765.11220061584, "data_wait_s": 1.8253998860018328e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203800, "loss": 0.5466072801500559, "lr": 0.0003, "grad_norm": 0.14604799335517452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29828659819, "tokens_per_second": 146933.57037424322, "data_wait_s": 2.0323001081123948e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203900, "loss": 0.5432954356074333, "lr": 0.0003, "grad_norm": 0.15455388367935946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29843379167, "tokens_per_second": 147760.30112497802, "data_wait_s": 1.8421000277157873e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 204000, "loss": 0.6074925512075424, "lr": 0.0003, "grad_norm": 0.14996113502412647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29858060304, "tokens_per_second": 148552.08069573547, "data_wait_s": 1.8466000256012194e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a020733d9a8808d2dbf0ac47da0750b59c81705a13ec91d06f402b90b62ecad5
-size 189450
+oid sha256:acac5f057d139a909695c0d77a71da9a35a26836dd87d62baf018635b4ca4c2e
+size 191568