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": 77019.45302047206,
-      "required_pages_per_second_upper_estimate": 0.005328384469708664
+      "required_token_positions_per_second": 76684.59750049555,
+      "required_pages_per_second_upper_estimate": 0.005305218387864837
     },
     {
       "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": 9627.431627559008,
-      "required_pages_per_second_upper_estimate": 0.0040308045788251395
+      "required_token_positions_per_second": 9585.574687561944,
+      "required_pages_per_second_upper_estimate": 0.00401327995212067
     },
     {
       "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": 19254.863255118016,
-      "required_pages_per_second_upper_estimate": 0.00401028023114999
+      "required_token_positions_per_second": 19171.14937512389,
+      "required_pages_per_second_upper_estimate": 0.003992844837630689
     },
     {
       "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": 28882.294882677023,
-      "required_pages_per_second_upper_estimate": 0.0017314494775057157
+      "required_token_positions_per_second": 28756.724062685833,
+      "required_pages_per_second_upper_estimate": 0.001723921698582784
     },
     {
       "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": 28882.294882677023,
-      "required_pages_per_second_upper_estimate": 0.0022454713873389285
+      "required_token_positions_per_second": 28756.724062685833,
+      "required_pages_per_second_upper_estimate": 0.002235708808412279
     },
     {
       "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": 28882.294882677023,
-      "required_pages_per_second_upper_estimate": 0.37047106736287405
+      "required_token_positions_per_second": 28756.724062685833,
+      "required_pages_per_second_upper_estimate": 0.36886037971146896
     }
   ],
   "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": 0.08472688012489016,
-    "producer_p95_seconds": 0.2978051689970016,
-    "gpu_update_seconds": 0.9572646533909144,
-    "producer_batches_per_second": 11.802629797367349,
-    "gpu_batches_per_second": 1.0446431887542327,
-    "producer_headroom": 11.298240322078133,
+    "producer_batch_seconds": 0.11355689749208864,
+    "producer_p95_seconds": 0.2948557579984481,
+    "gpu_update_seconds": 0.9614447021062287,
+    "producer_batches_per_second": 8.806158164629927,
+    "gpu_batches_per_second": 1.0401014200913568,
+    "producer_headroom": 8.466634113292953,
     "sustainable": true,
-    "buffer_seconds": 60.307673163627605,
+    "buffer_seconds": 60.57101623269241,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374605.0067519479,
+    "generation_bytes_per_second": 489647.7683797693,
     "upload_bytes_per_second": 4404232.847948614,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.38151409071759,
+    "estimated_full_shards_per_hour": 105.06701267762001,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -321,3 +321,4 @@
 {"step": 874600, "loss": 0.5458735823631287, "lr": 3e-05, "grad_norm": 0.32652412468524605, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128016788544, "tokens_per_second": 152586.37846397923, "data_wait_s": 1.3966997357783839e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 874700, "loss": 0.5483282804489136, "lr": 3e-05, "grad_norm": 0.3033618618893559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128031343974, "tokens_per_second": 151158.88541125078, "data_wait_s": 8.822000381769612e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 874800, "loss": 0.5186391472816467, "lr": 3e-05, "grad_norm": 0.3152294704003549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128046110866, "tokens_per_second": 152940.05195945117, "data_wait_s": 1.3041997590335086e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 874900, "loss": 0.537115216255188, "lr": 3e-05, "grad_norm": 0.30592999147303934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128060867543, "tokens_per_second": 154272.42569300413, "data_wait_s": 8.540002454537898e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd0c6b9d3884f7be10f3944def381036aa464d484bee7c561e8bd1ceb346035b
-size 688977
+oid sha256:1649145c851056b707fd4c50095063816c73b211f6b2f48f887c40b13eb28cfe
+size 691095