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": 75730.32128852022,
-      "required_pages_per_second_upper_estimate": 0.006564112312544398
+      "required_token_positions_per_second": 74695.32036647602,
+      "required_pages_per_second_upper_estimate": 0.006474401055807461
     },
     {
       "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": 9466.290161065028,
-      "required_pages_per_second_upper_estimate": 0.003776244678899405
+      "required_token_positions_per_second": 9336.915045809503,
+      "required_pages_per_second_upper_estimate": 0.0037246350110936265
     },
     {
       "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": 18932.580322130056,
-      "required_pages_per_second_upper_estimate": 0.004379719255491381
+      "required_token_positions_per_second": 18673.830091619006,
+      "required_pages_per_second_upper_estimate": 0.004319861943511174
     },
     {
       "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": 28398.870483195085,
-      "required_pages_per_second_upper_estimate": 0.0017293422150682779
+      "required_token_positions_per_second": 28010.74513742851,
+      "required_pages_per_second_upper_estimate": 0.0017057074178474063
     },
     {
       "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": 28398.870483195085,
-      "required_pages_per_second_upper_estimate": 0.0021862236629013178
+      "required_token_positions_per_second": 28010.74513742851,
+      "required_pages_per_second_upper_estimate": 0.002156344699384485
     },
     {
       "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": 28398.870483195085,
-      "required_pages_per_second_upper_estimate": 0.35261892649583526
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28010.74513742851,
+      "required_pages_per_second_upper_estimate": 0.3477997086733863
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 690670,
-    "producer_batch_seconds": 0.09338540646091786,
-    "producer_p95_seconds": 0.18998536500021146,
-    "gpu_update_seconds": 0.9735598469087209,
-    "producer_batches_per_second": 10.708311265085126,
-    "gpu_batches_per_second": 1.027158220601674,
-    "producer_headroom": 10.425181875887207,
+    "producer_batch_seconds": 0.5288739470781607,
+    "producer_p95_seconds": 0.19482192600025883,
+    "gpu_update_seconds": 0.9870497862285069,
+    "producer_batches_per_second": 1.8908097203967829,
+    "gpu_batches_per_second": 1.0131201221581492,
+    "producer_headroom": 1.8663233303164275,
     "sustainable": true,
-    "buffer_seconds": 60.360710508340695,
+    "buffer_seconds": 60.210036959938925,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 426186.01882479916,
-    "upload_bytes_per_second": 3066870.8165609036,
+    "generation_bytes_per_second": 369753.79488652706,
+    "upload_bytes_per_second": 3110855.373750247,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.44959853704435,
+    "estimated_full_shards_per_hour": 79.34055695483073,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 1218142,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.1807161469996572,
+      "bytes": 7525394,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -51,3 +51,4 @@
 {"step": 242200, "loss": 0.6197372898459435, "lr": 0.0003, "grad_norm": 0.1444105900824176, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35450713692, "tokens_per_second": 150344.98341179563, "data_wait_s": 1.4904000636306591e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242300, "loss": 0.6313554421067238, "lr": 0.0003, "grad_norm": 0.158225694374267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35465399501, "tokens_per_second": 149949.53320490653, "data_wait_s": 1.2076000530214515e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242400, "loss": 0.5764707215130329, "lr": 0.0003, "grad_norm": 0.1436426621896004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35479938250, "tokens_per_second": 149285.01365695783, "data_wait_s": 1.104199964174768e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 242500, "loss": 0.5573337953537703, "lr": 0.0003, "grad_norm": 0.14652808402817707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35494473423, "tokens_per_second": 146812.21251679005, "data_wait_s": 1.537099979032064e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:156e6acc847f4e6390807c2081a4d276a2f9a1c8f40a1627ed665cbd9e7f2f4d
-size 113202
+oid sha256:61e62a2f628d5865b95e8d6474806800d1883786cf48882ad904c140b7740952
+size 115320