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": 75966.11604179823,
-      "required_pages_per_second_upper_estimate": 0.005255512173445489
+      "required_token_positions_per_second": 76814.02817439828,
+      "required_pages_per_second_upper_estimate": 0.005314172702205967
     },
     {
       "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": 9495.764505224779,
-      "required_pages_per_second_upper_estimate": 0.003975678304225971
+      "required_token_positions_per_second": 9601.753521799785,
+      "required_pages_per_second_upper_estimate": 0.0040200536921635764
     },
     {
       "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": 18991.529010449558,
-      "required_pages_per_second_upper_estimate": 0.003955434652576586
+      "required_token_positions_per_second": 19203.50704359957,
+      "required_pages_per_second_upper_estimate": 0.003999584086645072
     },
     {
       "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": 28487.293515674333,
-      "required_pages_per_second_upper_estimate": 0.0017077697486860693
+      "required_token_positions_per_second": 28805.260565399352,
+      "required_pages_per_second_upper_estimate": 0.0017268313878095111
     },
     {
       "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": 28487.293515674333,
-      "required_pages_per_second_upper_estimate": 0.002214761768481866
+      "required_token_positions_per_second": 28805.260565399352,
+      "required_pages_per_second_upper_estimate": 0.002239482308008742
     },
     {
       "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": 28487.293515674333,
-      "required_pages_per_second_upper_estimate": 0.36540441394638773
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28805.260565399352,
+      "required_pages_per_second_upper_estimate": 0.36948295385384167
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.4452619788361005,
-    "producer_p95_seconds": 0.2617280470003607,
-    "gpu_update_seconds": 0.970537969315599,
-    "producer_batches_per_second": 2.2458688312304713,
-    "gpu_batches_per_second": 1.0303563916259526,
-    "producer_headroom": 2.1797009748116194,
+    "producer_batch_seconds": 0.08833682442170243,
+    "producer_p95_seconds": 0.2469352219995926,
+    "gpu_update_seconds": 0.959824679843742,
+    "producer_batches_per_second": 11.320307318567384,
+    "gpu_batches_per_second": 1.041856935959178,
+    "producer_headroom": 10.865510347776707,
     "sustainable": true,
-    "buffer_seconds": 60.173354097567135,
+    "buffer_seconds": 60.468954830155745,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331245.7009540533,
+    "generation_bytes_per_second": 330707.8441307348,
     "upload_bytes_per_second": 4907760.828050056,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.07761641946983,
+    "estimated_full_shards_per_hour": 70.96220486585172,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -228,3 +228,4 @@
 {"step": 419900, "loss": 0.6178654432296753, "lr": 0.0003, "grad_norm": 0.1400032057203984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61462422501, "tokens_per_second": 153448.85455168787, "data_wait_s": 1.3737000699620694e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420000, "loss": 0.6040761470794678, "lr": 0.0003, "grad_norm": 0.1615146686793636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61477032401, "tokens_per_second": 145438.4825535189, "data_wait_s": 8.343002264155075e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 420100, "loss": 0.5920270085334778, "lr": 0.0003, "grad_norm": 0.14548719408145425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61491509441, "tokens_per_second": 137382.28344275174, "data_wait_s": 1.4730001566931605e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 420200, "loss": 0.5941126346588135, "lr": 0.0003, "grad_norm": 0.181814914446555, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61506083731, "tokens_per_second": 150840.83369612007, "data_wait_s": 1.7055001080734655e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c81e8740d1fda3d788185e2ccf0d39c7d814957db107c7640ffe5566e296223d
-size 491220
+oid sha256:5d26ef9c7fa34e0c56dd6263838110ab261c361704602128bbe2f21ff6cee2ec
+size 493338