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": 75648.93950886454,
-      "required_pages_per_second_upper_estimate": 0.006557058346144078
+      "required_token_positions_per_second": 73539.62407958538,
+      "required_pages_per_second_upper_estimate": 0.006374228230745241
     },
     {
       "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": 9456.117438608067,
-      "required_pages_per_second_upper_estimate": 0.0037721866278155685
+      "required_token_positions_per_second": 9192.453009948173,
+      "required_pages_per_second_upper_estimate": 0.003667006945088628
     },
     {
       "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": 18912.234877216135,
-      "required_pages_per_second_upper_estimate": 0.0043750126948781465
+      "required_token_positions_per_second": 18384.906019896345,
+      "required_pages_per_second_upper_estimate": 0.004253024444408128
     },
     {
       "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": 28368.352315824206,
-      "required_pages_per_second_upper_estimate": 0.0017274838187918411
+      "required_token_positions_per_second": 27577.359029844516,
+      "required_pages_per_second_upper_estimate": 0.0016793164776967191
     },
     {
       "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": 28368.352315824206,
-      "required_pages_per_second_upper_estimate": 0.002183874289897413
+      "required_token_positions_per_second": 27577.359029844516,
+      "required_pages_per_second_upper_estimate": 0.00212298143713314
     },
     {
       "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": 28368.352315824206,
-      "required_pages_per_second_upper_estimate": 0.35223999299482484
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27577.359029844516,
+      "required_pages_per_second_upper_estimate": 0.34241850366719045
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.061839020335966666,
-    "producer_p95_seconds": 0.17076587500014284,
-    "gpu_update_seconds": 0.974607185225122,
-    "producer_batches_per_second": 16.17101943994385,
-    "gpu_batches_per_second": 1.0260544095711879,
-    "producer_headroom": 15.760391738584405,
+    "producer_batch_seconds": 0.32265425971874606,
+    "producer_p95_seconds": 0.17665724600010435,
+    "gpu_update_seconds": 1.002561556749471,
+    "producer_batches_per_second": 3.099292725506517,
+    "gpu_batches_per_second": 0.9974449880586125,
+    "producer_headroom": 3.1072317397061244,
     "sustainable": true,
-    "buffer_seconds": 60.42564548395757,
+    "buffer_seconds": 60.15369340496826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 140109.69730787992,
+    "upload_bytes_per_second": 3580018.611091431,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 30.064279455445273,
     "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": 1.272373552999852,
+      "bytes": 4555121,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -46,3 +46,4 @@
 {"step": 119600, "loss": 0.6463970169425011, "lr": 0.0003, "grad_norm": 0.15078117921742035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17502101396, "tokens_per_second": 150354.02502220837, "data_wait_s": 9.141999726125505e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 119700, "loss": 0.6455818898975849, "lr": 0.0003, "grad_norm": 0.13960348146462503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17516653535, "tokens_per_second": 149283.06221811823, "data_wait_s": 1.7284000023209956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 119800, "loss": 0.5821646936237812, "lr": 0.0003, "grad_norm": 0.15459538561565184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17531207737, "tokens_per_second": 148681.6515178564, "data_wait_s": 1.4895000276737846e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 119900, "loss": 0.574279710650444, "lr": 0.0003, "grad_norm": 0.1383865787803532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 17545806061, "tokens_per_second": 148313.0207369725, "data_wait_s": 1.1406999874452595e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6620b62c3e609f5d54d0c17a6762d533e2bd0837890f81a06f750e8ba031cc9
-size 101829
+oid sha256:607dfb879720aa33aa31eff69ed54b107de5b0edff4d1778ee7fb863429adb1d
+size 103947