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": 75876.2563471976,
-      "required_pages_per_second_upper_estimate": 0.0052492954712698
+      "required_token_positions_per_second": 77558.12125342256,
+      "required_pages_per_second_upper_estimate": 0.00536565078794668
     },
     {
       "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": 9484.5320433997,
-      "required_pages_per_second_upper_estimate": 0.003970975507020286
+      "required_token_positions_per_second": 9694.76515667782,
+      "required_pages_per_second_upper_estimate": 0.004058995721383207
     },
     {
       "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": 18969.0640867994,
-      "required_pages_per_second_upper_estimate": 0.0039507558014201345
+      "required_token_positions_per_second": 19389.53031335564,
+      "required_pages_per_second_upper_estimate": 0.00403832782797174
     },
     {
       "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": 28453.5961301991,
-      "required_pages_per_second_upper_estimate": 0.0017057496418797576
+      "required_token_positions_per_second": 29084.295470033463,
+      "required_pages_per_second_upper_estimate": 0.0017435591042806965
     },
     {
       "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": 28453.5961301991,
-      "required_pages_per_second_upper_estimate": 0.002212141944980301
+      "required_token_positions_per_second": 29084.295470033463,
+      "required_pages_per_second_upper_estimate": 0.0022611760444992083
     },
     {
       "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": 28453.5961301991,
-      "required_pages_per_second_upper_estimate": 0.36497218006694504
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29084.295470033463,
+      "required_pages_per_second_upper_estimate": 0.37306211400615
     }
   ],
   "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": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.5087933665076321,
-    "producer_p95_seconds": 0.3344976200023666,
-    "gpu_update_seconds": 0.9716873703234972,
-    "producer_batches_per_second": 1.965434429430596,
-    "gpu_batches_per_second": 1.0291375915147245,
-    "producer_headroom": 1.909787812276679,
+    "producer_batch_seconds": 0.16870178959356963,
+    "producer_p95_seconds": 0.33358344699809095,
+    "gpu_update_seconds": 0.9506161161265424,
+    "producer_batches_per_second": 5.927619395201228,
+    "gpu_batches_per_second": 1.051949344257576,
+    "producer_headroom": 5.634890527342556,
     "sustainable": true,
-    "buffer_seconds": 60.244616960056824,
+    "buffer_seconds": 60.83943143209871,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 252733.28326514678,
+    "generation_bytes_per_second": 259497.12395813738,
     "upload_bytes_per_second": 5162767.430630665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.2306792589741,
+    "estimated_full_shards_per_hour": 55.68204201753703,
     "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

@@ -404,3 +404,4 @@
 {"step": 925400, "loss": 0.5439279079437256, "lr": 3e-05, "grad_norm": 0.3367022718858882, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135452275707, "tokens_per_second": 153383.2879145928, "data_wait_s": 1.4245000784285367e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 925500, "loss": 0.462994247674942, "lr": 3e-05, "grad_norm": 0.33712264966163574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135466831988, "tokens_per_second": 140100.0005896525, "data_wait_s": 1.8733000615611672e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 925600, "loss": 0.5387578010559082, "lr": 3e-05, "grad_norm": 0.33788941470807277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135481383545, "tokens_per_second": 151936.79056625575, "data_wait_s": 7.653005013708025e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 925700, "loss": 0.5607717037200928, "lr": 3e-05, "grad_norm": 0.30635413400909806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135496029460, "tokens_per_second": 153080.14015474648, "data_wait_s": 1.4951998309697956e-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:f5b0db00ce82cb493a6e8ab29752a165f8b56df979461a4d903a1a8a4f8dfc53
-size 866337
+oid sha256:a2cdf944bf93299184260021ad43b716686d357afe662dac262571c29f19f29e
+size 868455