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": 76049.57910491506,
-      "required_pages_per_second_upper_estimate": 0.005261286341812899
+      "required_token_positions_per_second": 76432.37212193751,
+      "required_pages_per_second_upper_estimate": 0.005287768850932683
     },
     {
       "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": 9506.197388114382,
-      "required_pages_per_second_upper_estimate": 0.003980046334428479
+      "required_token_positions_per_second": 9554.04651524219,
+      "required_pages_per_second_upper_estimate": 0.004000079764753494
     },
     {
       "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": 19012.394776228764,
-      "required_pages_per_second_upper_estimate": 0.003959780441321148
+      "required_token_positions_per_second": 19108.09303048438,
+      "required_pages_per_second_upper_estimate": 0.003979711863949913
     },
     {
       "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": 28518.592164343147,
-      "required_pages_per_second_upper_estimate": 0.0017096460548834953
+      "required_token_positions_per_second": 28662.139545726568,
+      "required_pages_per_second_upper_estimate": 0.0017182515012132713
     },
     {
       "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": 28518.592164343147,
-      "required_pages_per_second_upper_estimate": 0.0022171951007476597
+      "required_token_positions_per_second": 28662.139545726568,
+      "required_pages_per_second_upper_estimate": 0.0022283552782520182
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28518.592164343147,
-      "required_pages_per_second_upper_estimate": 0.3658058794056406
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28662.139545726568,
+      "required_pages_per_second_upper_estimate": 0.3676471510848574
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08296419136689792,
-    "producer_p95_seconds": 0.2764361720001034,
-    "gpu_update_seconds": 0.9694728211222274,
-    "producer_batches_per_second": 12.053392958145462,
-    "gpu_batches_per_second": 1.0314884318700501,
-    "producer_headroom": 11.685436875228072,
+    "producer_batch_seconds": 0.10079050995310013,
+    "producer_p95_seconds": 0.2933985159979784,
+    "gpu_update_seconds": 0.9646174513905829,
+    "producer_batches_per_second": 9.921569009476391,
+    "gpu_batches_per_second": 1.0366803944490222,
+    "producer_headroom": 9.570518611716906,
     "sustainable": true,
-    "buffer_seconds": 60.1073149095781,
+    "buffer_seconds": 60.77089943760672,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 499040.6279768092,
+    "generation_bytes_per_second": 502711.4004729321,
     "upload_bytes_per_second": 4682397.744684979,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.08250169256408,
+    "estimated_full_shards_per_hour": 107.87016401902171,
     "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-v1/1791399463488568880-054ba646/train.jsonl

@@ -318,3 +318,4 @@
 {"step": 916800, "loss": 0.5456134676933289, "lr": 3e-05, "grad_norm": 0.3340009759926441, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134193216992, "tokens_per_second": 152473.58338664877, "data_wait_s": 1.3147000572644174e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 916900, "loss": 0.5371228456497192, "lr": 3e-05, "grad_norm": 0.3320695904305587, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134207749818, "tokens_per_second": 151025.80131240535, "data_wait_s": 1.263399826711975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917000, "loss": 0.5400528907775879, "lr": 3e-05, "grad_norm": 0.3300478777332138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134222322698, "tokens_per_second": 152410.60590289687, "data_wait_s": 9.42399856285192e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 917100, "loss": 0.5253779292106628, "lr": 3e-05, "grad_norm": 0.33882088372326946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134236869538, "tokens_per_second": 152280.56928784386, "data_wait_s": 1.7545000446261838e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:27258a8de26dfbe168d9d5b6b17b903df2034d995976078121ea4e1a4f777b75
-size 682623
+oid sha256:9080a9bccd8c316282f0bd83778c47d3bea41c3fa4a4f544a0ecefb245ed6b85
+size 684741