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": 75727.80439770009,
-      "required_pages_per_second_upper_estimate": 0.005239025220947583
+      "required_token_positions_per_second": 76976.4288624453,
+      "required_pages_per_second_upper_estimate": 0.005325407958626572
     },
     {
       "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": 9465.975549712512,
-      "required_pages_per_second_upper_estimate": 0.003963206290617112
+      "required_token_positions_per_second": 9622.053607805663,
+      "required_pages_per_second_upper_estimate": 0.004028552914260237
     },
     {
       "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": 18931.951099425023,
-      "required_pages_per_second_upper_estimate": 0.003943026144885346
+      "required_token_positions_per_second": 19244.107215611326,
+      "required_pages_per_second_upper_estimate": 0.004008040031776584
     },
     {
       "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": 28397.926649137535,
-      "required_pages_per_second_upper_estimate": 0.0017024123414924924
+      "required_token_positions_per_second": 28866.16082341699,
+      "required_pages_per_second_upper_estimate": 0.0017304822652883588
     },
     {
       "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": 28397.926649137535,
-      "required_pages_per_second_upper_estimate": 0.0022078138876919333
+      "required_token_positions_per_second": 28866.16082341699,
+      "required_pages_per_second_upper_estimate": 0.002244217035197688
     },
     {
       "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": 28397.926649137535,
-      "required_pages_per_second_upper_estimate": 0.36425811173711903
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28866.16082341699,
+      "required_pages_per_second_upper_estimate": 0.3702641169740895
     }
   ],
   "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": 0.2874758385001712,
-    "producer_p95_seconds": 0.25645213200186845,
-    "gpu_update_seconds": 0.9735922041632462,
-    "producer_batches_per_second": 3.4785532071746768,
-    "gpu_batches_per_second": 1.027124083085125,
-    "producer_headroom": 3.3866922842723226,
+    "producer_batch_seconds": 0.42135996978916523,
+    "producer_p95_seconds": 0.2509879589997581,
+    "gpu_update_seconds": 0.9577996938744696,
+    "producer_batches_per_second": 2.373267684873737,
+    "gpu_batches_per_second": 1.0440596362636354,
+    "producer_headroom": 2.2731150620542366,
     "sustainable": true,
-    "buffer_seconds": 60.362716658121265,
+    "buffer_seconds": 60.341380714091585,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 441743.559760482,
+    "generation_bytes_per_second": 381730.89849312016,
     "upload_bytes_per_second": 5189913.752652345,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.78788466082425,
+    "estimated_full_shards_per_hour": 81.91056457610324,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -168,3 +168,4 @@
 {"step": 776800, "loss": 0.5557436943054199, "lr": 3e-05, "grad_norm": 0.2563867475895094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113701554340, "tokens_per_second": 153640.40915584387, "data_wait_s": 9.08000220078975e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 776900, "loss": 0.5226836800575256, "lr": 3e-05, "grad_norm": 0.26258257862796947, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113716248633, "tokens_per_second": 151190.80186561932, "data_wait_s": 1.3761000445811078e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 777000, "loss": 0.52857905626297, "lr": 3e-05, "grad_norm": 0.26690178068427073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113730736548, "tokens_per_second": 150901.2304228069, "data_wait_s": 9.418003173777834e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 777100, "loss": 0.5477849245071411, "lr": 3e-05, "grad_norm": 0.25391972160481335, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113745248844, "tokens_per_second": 150393.84277051754, "data_wait_s": 7.638001989107579e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f85be2d7e63a6e7135973e514b5ef7dff869724a2fabba8bfa0705357908d359
-size 362574
+oid sha256:9b662a726cfb82133f81cd2c54d5f3f44f86d64cd69f10a707d665c8ef37d3bd
+size 364692