Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75180.14754977364,
-      "required_pages_per_second_upper_estimate": 0.006516424647272437
+      "required_token_positions_per_second": 74031.32072402074,
+      "required_pages_per_second_upper_estimate": 0.006416847249691146
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9397.518443721705,
-      "required_pages_per_second_upper_estimate": 0.003748810612622349
+      "required_token_positions_per_second": 9253.915090502593,
+      "required_pages_per_second_upper_estimate": 0.003691525087961781
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18795.03688744341,
-      "required_pages_per_second_upper_estimate": 0.0043479010025584
+      "required_token_positions_per_second": 18507.830181005185,
+      "required_pages_per_second_upper_estimate": 0.004281460785689311
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28192.555331165113,
-      "required_pages_per_second_upper_estimate": 0.001716778704761599
+      "required_token_positions_per_second": 27761.745271507778,
+      "required_pages_per_second_upper_estimate": 0.0016905446324141664
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28192.555331165113,
-      "required_pages_per_second_upper_estimate": 0.0021703409513811564
+      "required_token_positions_per_second": 27761.745271507778,
+      "required_pages_per_second_upper_estimate": 0.0021371760004301626
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28192.555331165113,
-      "required_pages_per_second_upper_estimate": 0.3500571828000188
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27761.745271507778,
+      "required_pages_per_second_upper_estimate": 0.34470796368759427
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.061892193023446396,
-    "producer_p95_seconds": 0.15954064400011703,
-    "gpu_update_seconds": 0.9806844280424931,
-    "producer_batches_per_second": 16.15712662857452,
-    "gpu_batches_per_second": 1.019696011688553,
-    "producer_headroom": 15.84504248655374,
+    "producer_batch_seconds": 0.04527427875000711,
+    "producer_p95_seconds": 0.1502281419998326,
+    "gpu_update_seconds": 0.995902805447015,
+    "producer_batches_per_second": 22.08759648103379,
+    "gpu_batches_per_second": 1.004114050618771,
+    "producer_headroom": 21.997099301043168,
     "sustainable": true,
-    "buffer_seconds": 60.80243453863457,
+    "buffer_seconds": 60.75007113226792,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 332718.32522702555,
+    "upload_bytes_per_second": 18634.022848171087,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 71.39360730751109,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 1.3317575169999145,
+      "bytes": 24816,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -28,3 +28,4 @@
 {"step": 197800, "loss": 0.5235539898276329, "lr": 0.0003, "grad_norm": 0.149242034860401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28949997669, "tokens_per_second": 150145.07525097489, "data_wait_s": 1.1808000181190437e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 197900, "loss": 0.5999762620776892, "lr": 0.0003, "grad_norm": 0.14391978348168138, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28964748234, "tokens_per_second": 150920.02659706326, "data_wait_s": 1.72149998434179e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 198000, "loss": 0.6337952017784119, "lr": 0.0003, "grad_norm": 0.18071502585798865, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28979500934, "tokens_per_second": 149136.57135252244, "data_wait_s": 1.6908999896259047e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 198100, "loss": 0.5864313617348671, "lr": 0.0003, "grad_norm": 0.13070283264529284, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28994197011, "tokens_per_second": 148100.4656571597, "data_wait_s": 1.8413000361761078e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:74794e4cb4f5b729e458198a06fe1e932e1114f9e37ae3e97cb0b55e32f651ea
-size 63705
+oid sha256:99271ae9f490ddd44245ecc236051bfa4eae73edc7bb840b2789d774453b97e5
+size 65823