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": 77114.40389881088,
-      "required_pages_per_second_upper_estimate": 0.005334953391788535
+      "required_token_positions_per_second": 77115.79531992723,
+      "required_pages_per_second_upper_estimate": 0.005335049653529905
     },
     {
       "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": 9639.30048735136,
-      "required_pages_per_second_upper_estimate": 0.004035773822570221
+      "required_token_positions_per_second": 9639.474414990904,
+      "required_pages_per_second_upper_estimate": 0.004035846642440876
     },
     {
       "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": 19278.60097470272,
-      "required_pages_per_second_upper_estimate": 0.004015224172133722
+      "required_token_positions_per_second": 19278.948829981808,
+      "required_pages_per_second_upper_estimate": 0.004015296621214795
     },
     {
       "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": 28917.90146205408,
-      "required_pages_per_second_upper_estimate": 0.0017335840375712717
+      "required_token_positions_per_second": 28918.423244972713,
+      "required_pages_per_second_upper_estimate": 0.0017336153176605272
     },
     {
       "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": 28917.90146205408,
-      "required_pages_per_second_upper_estimate": 0.002248239642268704
+      "required_token_positions_per_second": 28918.423244972713,
+      "required_pages_per_second_upper_estimate": 0.002248280208595547
     },
     {
       "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": 28917.90146205408,
-      "required_pages_per_second_upper_estimate": 0.37092779033175666
+      "required_token_positions_per_second": 28918.423244972713,
+      "required_pages_per_second_upper_estimate": 0.37093448320279004
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07902154340621337,
-    "producer_p95_seconds": 0.20074726500024553,
-    "gpu_update_seconds": 0.9560859745054309,
-    "producer_batches_per_second": 12.654776873433875,
-    "gpu_batches_per_second": 1.0459310424643402,
-    "producer_headroom": 12.099054679185816,
+    "producer_batch_seconds": 0.343836432000046,
+    "producer_p95_seconds": 0.22972871499950998,
+    "gpu_update_seconds": 0.9560687235880481,
+    "producer_batches_per_second": 2.908359635374143,
+    "gpu_batches_per_second": 1.0459499148210616,
+    "producer_headroom": 2.7805916843271574,
     "sustainable": true,
-    "buffer_seconds": 60.23341639384215,
+    "buffer_seconds": 60.23232958604703,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 171,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 673494.1134022513,
+    "generation_bytes_per_second": 599114.8376553796,
     "upload_bytes_per_second": 5275140.440368012,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.2580792977841,
+    "estimated_full_shards_per_hour": 64.2780487406065,
     "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

@@ -103,3 +103,4 @@
 {"step": 895300, "loss": 0.5230830907821655, "lr": 3e-05, "grad_norm": 0.3115013720633862, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131046724901, "tokens_per_second": 153471.71832988693, "data_wait_s": 3.731300057552289e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895400, "loss": 0.5063887238502502, "lr": 3e-05, "grad_norm": 0.325257508165069, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131061408234, "tokens_per_second": 153339.8873856963, "data_wait_s": 1.215900010720361e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895500, "loss": 0.5193361639976501, "lr": 3e-05, "grad_norm": 0.3058153376724782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131076165805, "tokens_per_second": 153918.2519494799, "data_wait_s": 2.2959000489208847e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 895600, "loss": 0.5518725514411926, "lr": 3e-05, "grad_norm": 0.3277005091052649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131090877456, "tokens_per_second": 153869.18445872568, "data_wait_s": 1.3012999261263758e-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:d0a0634e351bd966877651895e616ab62aa866de9ceba4925600fe0d3eca1ecd
-size 224121
+oid sha256:aa0de1bfa0bc1a2f6f83689c7cd3b9cde2cffcaa01008a84d4966c9f25ff17e9
+size 226239