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": 75606.06524473379,
-      "required_pages_per_second_upper_estimate": 0.006553342113593054
+      "required_token_positions_per_second": 76362.50788526027,
+      "required_pages_per_second_upper_estimate": 0.006618908644487545
     },
     {
       "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": 9450.758155591724,
-      "required_pages_per_second_upper_estimate": 0.0037700487296919278
+      "required_token_positions_per_second": 9545.313485657534,
+      "required_pages_per_second_upper_estimate": 0.003807768264583347
     },
     {
       "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": 18901.516311183448,
-      "required_pages_per_second_upper_estimate": 0.004372533143266804
+      "required_token_positions_per_second": 19090.62697131507,
+      "required_pages_per_second_upper_estimate": 0.00441628056625431
     },
     {
       "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": 28352.274466775172,
-      "required_pages_per_second_upper_estimate": 0.0017265047621387098
+      "required_token_positions_per_second": 28635.940456972603,
+      "required_pages_per_second_upper_estimate": 0.0017437785326613032
     },
     {
       "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": 28352.274466775172,
-      "required_pages_per_second_upper_estimate": 0.0021826365725712837
+      "required_token_positions_per_second": 28635.940456972603,
+      "required_pages_per_second_upper_estimate": 0.0022044739657344006
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28352.274466775172,
-      "required_pages_per_second_upper_estimate": 0.35204035991873517
+      "required_token_positions_per_second": 28635.940456972603,
+      "required_pages_per_second_upper_estimate": 0.35556254214798916
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.1390997315779714,
-    "producer_p95_seconds": 0.3573999280051794,
-    "gpu_update_seconds": 0.9751598600105088,
-    "producer_batches_per_second": 7.189086482452749,
-    "gpu_batches_per_second": 1.0254728901466714,
-    "producer_headroom": 7.010508567832064,
+    "producer_batch_seconds": 0.13634472578132772,
+    "producer_p95_seconds": 0.3520125119976001,
+    "gpu_update_seconds": 0.9654999821480614,
+    "producer_batches_per_second": 7.334350443476773,
+    "gpu_batches_per_second": 1.0357328000930486,
+    "producer_headroom": 7.08131522224445,
     "sustainable": true,
-    "buffer_seconds": 60.45991132065154,
+    "buffer_seconds": 60.82649887532787,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 261882.01686653527,
-    "upload_bytes_per_second": 5607397.059808457,
+    "generation_bytes_per_second": 269709.77910281264,
+    "upload_bytes_per_second": 5626744.633637237,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.19378451821369,
+    "estimated_full_shards_per_hour": 57.873440073139996,
     "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": [
-    {
-      "seconds": 1.1613783780012454,
-      "bytes": 6337641,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.4139873040003295,
       "bytes": 5988516,
@@ -562,6 +556,12 @@
       "bytes": 830736,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1426053430041065,
+      "bytes": 7670085,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -279,3 +279,4 @@
 {"step": 308100, "loss": 0.5738916657865047, "lr": 0.0003, "grad_norm": 0.15274501562727477, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45097640050, "tokens_per_second": 151146.1361488563, "data_wait_s": 1.0962998203467578e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 308200, "loss": 0.6042536422610283, "lr": 0.0003, "grad_norm": 0.15310585430395826, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45112381171, "tokens_per_second": 151511.93431939805, "data_wait_s": 9.064999176189303e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 308300, "loss": 0.5641783438622952, "lr": 0.0003, "grad_norm": 0.18553824204848238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45126962412, "tokens_per_second": 149157.4715018199, "data_wait_s": 9.802002750802785e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 308400, "loss": 0.5799893271178007, "lr": 0.0003, "grad_norm": 0.16008279980643045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45141484024, "tokens_per_second": 148829.09510379934, "data_wait_s": 1.3167998986318707e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:35b30fbdec74978a44db7ed785ab695e16cd4f2d0e1c4bc30daa698e54d8677d
-size 599238
+oid sha256:cbda4d0945c8db58d0e8e5fc0d4deb91a6661cbfaa4159c3e3f30b8d239589a3
+size 601356