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": 74874.81148344805,
-      "required_pages_per_second_upper_estimate": 0.00648995889091052
+      "required_token_positions_per_second": 75901.6656764869,
+      "required_pages_per_second_upper_estimate": 0.006578964009824969
     },
     {
       "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": 9359.351435431006,
-      "required_pages_per_second_upper_estimate": 0.003733585222367563
+      "required_token_positions_per_second": 9487.708209560862,
+      "required_pages_per_second_upper_estimate": 0.003784788658672755
     },
     {
       "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": 18718.702870862013,
-      "required_pages_per_second_upper_estimate": 0.0043302424712551016
+      "required_token_positions_per_second": 18975.416419121724,
+      "required_pages_per_second_upper_estimate": 0.0043896286326408455
     },
     {
       "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": 28078.054306293023,
-      "required_pages_per_second_upper_estimate": 0.0017098061930873386
+      "required_token_positions_per_second": 28463.124628682584,
+      "required_pages_per_second_upper_estimate": 0.001733254955413019
     },
     {
       "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": 28078.054306293023,
-      "required_pages_per_second_upper_estimate": 0.0021615263455273797
+      "required_token_positions_per_second": 28463.124628682584,
+      "required_pages_per_second_upper_estimate": 0.002191170124887804
     },
     {
       "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": 28078.054306293023,
-      "required_pages_per_second_upper_estimate": 0.3486354632813865
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28463.124628682584,
+      "required_pages_per_second_upper_estimate": 0.3534167479379985
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 690670,
-    "producer_batch_seconds": 0.1179113319373073,
-    "producer_p95_seconds": 0.3615771410004527,
-    "gpu_update_seconds": 0.9846836144128179,
-    "producer_batches_per_second": 8.48094906206041,
-    "gpu_batches_per_second": 1.0155546262403436,
-    "producer_headroom": 8.351051576080643,
+    "producer_batch_seconds": 0.14093382614831285,
+    "producer_p95_seconds": 0.35941645599814365,
+    "gpu_update_seconds": 0.9713620820160702,
+    "producer_batches_per_second": 7.095528641560061,
+    "gpu_batches_per_second": 1.0294822275999198,
+    "producer_headroom": 6.892327474270439,
     "sustainable": true,
-    "buffer_seconds": 60.06570047918189,
+    "buffer_seconds": 60.22444908499635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420023.04308641335,
-    "upload_bytes_per_second": 3188839.5755910934,
+    "generation_bytes_per_second": 436453.6801078162,
+    "upload_bytes_per_second": 3147383.6540707,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.12716741031933,
+    "estimated_full_shards_per_hour": 93.65279962945809,
     "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": [
-    {
-      "seconds": 1.146538889999647,
-      "bytes": 5754870,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 1.1586719139995694,
       "bytes": 7299186,
@@ -562,6 +556,12 @@
       "bytes": 6134154,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.9025495240020973,
+      "bytes": 2606249,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 346400, "loss": 0.5925719030201435, "lr": 0.0003, "grad_norm": 0.1437341434483271, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50703960052, "tokens_per_second": 147306.37665867622, "data_wait_s": 1.664400042500347e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346500, "loss": 0.622475216165185, "lr": 0.0003, "grad_norm": 0.1781314636102928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50718501158, "tokens_per_second": 148319.9760874813, "data_wait_s": 1.1948999599553645e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 346600, "loss": 0.595458822324872, "lr": 0.0003, "grad_norm": 0.15219270894335488, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50733088387, "tokens_per_second": 148684.9331796579, "data_wait_s": 1.6601999959675595e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 346700, "loss": 0.6234288327395916, "lr": 0.0003, "grad_norm": 0.1694497640742114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50747791494, "tokens_per_second": 138618.26169335644, "data_wait_s": 1.367900040349923e-05, "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 sha256:64fcf90e9bae32652eb62f023e51c4a049ec1097e3bed89e52d0ff8994cd3121
-size 674151
+oid sha256:04d16d81669b48c6f9f5572a95ec5642622068752a157f6bf5aa5ed4a1fc97d0
+size 676269