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": 76352.87076383106,
-      "required_pages_per_second_upper_estimate": 0.005282268762510312
+      "required_token_positions_per_second": 77441.85958919808,
+      "required_pages_per_second_upper_estimate": 0.005357607536251403
     },
     {
       "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": 9544.108845478882,
-      "required_pages_per_second_upper_estimate": 0.00399591906994574
+      "required_token_positions_per_second": 9680.23244864976,
+      "required_pages_per_second_upper_estimate": 0.0040529111800093115
     },
     {
       "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": 19088.217690957765,
-      "required_pages_per_second_upper_estimate": 0.00397557235487447
+      "required_token_positions_per_second": 19360.46489729952,
+      "required_pages_per_second_upper_estimate": 0.004032274268313817
     },
     {
       "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": 28632.32653643665,
-      "required_pages_per_second_upper_estimate": 0.00171646425682817
+      "required_token_positions_per_second": 29040.69734594928,
+      "required_pages_per_second_upper_estimate": 0.0017409454633123306
     },
     {
       "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": 28632.32653643665,
-      "required_pages_per_second_upper_estimate": 0.0022260374479132965
+      "required_token_positions_per_second": 29040.69734594928,
+      "required_pages_per_second_upper_estimate": 0.0022577864821195443
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28632.32653643665,
-      "required_pages_per_second_upper_estimate": 0.3672647418124017
+      "required_token_positions_per_second": 29040.69734594928,
+      "required_pages_per_second_upper_estimate": 0.3725028840824166
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08090288797649237,
-    "producer_p95_seconds": 0.29312131800179486,
-    "gpu_update_seconds": 0.9656218458117951,
-    "producer_batches_per_second": 12.360498185065607,
-    "gpu_batches_per_second": 1.035602088268108,
-    "producer_headroom": 11.935567072616395,
+    "producer_batch_seconds": 0.20072268035923457,
+    "producer_p95_seconds": 0.29584144100226695,
+    "gpu_update_seconds": 0.9520432540114765,
+    "producer_batches_per_second": 4.981998039336133,
+    "gpu_batches_per_second": 1.0503724445149478,
+    "producer_headroom": 4.743077624848368,
     "sustainable": true,
-    "buffer_seconds": 60.83417628614309,
+    "buffer_seconds": 60.930768256734495,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 166,
+    "upload_pages_per_commit": 158,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 572423.5475343943,
+    "generation_bytes_per_second": 570073.8837509342,
     "upload_bytes_per_second": 4780972.6909210365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.41438398134171,
+    "estimated_full_shards_per_hour": 61.162292406063166,
     "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

@@ -328,3 +328,4 @@
 {"step": 917800, "loss": 0.5411073565483093, "lr": 3e-05, "grad_norm": 0.33555589197663044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134339171440, "tokens_per_second": 152849.26160328538, "data_wait_s": 1.454599987482652e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 917900, "loss": 0.5300824046134949, "lr": 3e-05, "grad_norm": 0.33125594738717046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134353833516, "tokens_per_second": 152752.2748519518, "data_wait_s": 1.5485999028896913e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 918000, "loss": 0.5584174394607544, "lr": 3e-05, "grad_norm": 0.3468850130237999, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134368569035, "tokens_per_second": 153962.1086885535, "data_wait_s": 1.3860000763088465e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 918100, "loss": 0.5291762948036194, "lr": 3e-05, "grad_norm": 0.31930739252881585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134383307774, "tokens_per_second": 153637.1333065604, "data_wait_s": 1.6166999557754025e-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 sha256:0645b652fe9f863629f9b9eaa548fc4c29ef5e75d8b19d7a5c71398097269488
-size 703803
+oid sha256:5a84d77c86654c422f046775b5ca2b3714a5fb6f364526eda2de2be55674691a
+size 705921