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": 76228.78957534443,
-      "required_pages_per_second_upper_estimate": 0.006607318280289998
+      "required_token_positions_per_second": 75656.76892193005,
+      "required_pages_per_second_upper_estimate": 0.006557736979825131
     },
     {
       "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": 9528.598696918054,
-      "required_pages_per_second_upper_estimate": 0.0038011004854468064
+      "required_token_positions_per_second": 9457.096115241256,
+      "required_pages_per_second_upper_estimate": 0.0037725770365570673
     },
     {
       "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": 19057.197393836108,
-      "required_pages_per_second_upper_estimate": 0.004408547221845021
+      "required_token_positions_per_second": 18914.192230482513,
+      "required_pages_per_second_upper_estimate": 0.004375465494108083
     },
     {
       "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": 28585.79609075416,
-      "required_pages_per_second_upper_estimate": 0.001740725003845757
+      "required_token_positions_per_second": 28371.28834572377,
+      "required_pages_per_second_upper_estimate": 0.0017276626076085661
     },
     {
       "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": 28585.79609075416,
-      "required_pages_per_second_upper_estimate": 0.002200613713614415
+      "required_token_positions_per_second": 28371.28834572377,
+      "required_pages_per_second_upper_estimate": 0.0021841003136065327
     },
     {
       "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": 28585.79609075416,
-      "required_pages_per_second_upper_estimate": 0.35493991694195415
+      "required_token_positions_per_second": 28371.28834572377,
+      "required_pages_per_second_upper_estimate": 0.3522764486599174
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.09020478931182652,
+    "producer_batch_seconds": 0.11031103473442272,
     "producer_p95_seconds": 0.34602925700164633,
-    "gpu_update_seconds": 0.9671936339370488,
-    "producer_batches_per_second": 11.085885878444069,
-    "gpu_batches_per_second": 1.0339191294398928,
-    "producer_headroom": 10.722198248183732,
+    "gpu_update_seconds": 0.9745063270687605,
+    "producer_batches_per_second": 9.065276220167197,
+    "gpu_batches_per_second": 1.0261606027822543,
+    "producer_headroom": 8.834169033178911,
     "sustainable": true,
-    "buffer_seconds": 60.933198938034074,
+    "buffer_seconds": 60.41939227826315,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342705.57720936526,
-    "upload_bytes_per_second": 5431646.432991981,
+    "generation_bytes_per_second": 391527.36741346045,
+    "upload_bytes_per_second": 5269501.581547472,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.53663909159391,
+    "estimated_full_shards_per_hour": 84.01265875628339,
     "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": 1.2268210219990578,
-      "bytes": 15731128,
-      "pages": 39,
-      "shards": 4
-    },
     {
       "seconds": 0.8699136540017207,
       "bytes": 2665528,
@@ -562,6 +556,12 @@
       "bytes": 13743089,
       "pages": 49,
       "shards": 4
+    },
+    {
+      "seconds": 0.8212157870002557,
+      "bytes": 890133,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -377,3 +377,4 @@
 {"step": 392700, "loss": 0.5772739723324776, "lr": 0.0003, "grad_norm": 0.17186824175750987, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57480678216, "tokens_per_second": 148671.30602763983, "data_wait_s": 1.1964002624154091e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 392800, "loss": 0.5447807144373655, "lr": 0.0003, "grad_norm": 0.14101253288687018, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57495242606, "tokens_per_second": 149621.282252875, "data_wait_s": 1.1374999303370714e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 392900, "loss": 0.5725276041775942, "lr": 0.0003, "grad_norm": 0.15255747784283108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57509779999, "tokens_per_second": 147895.89224377097, "data_wait_s": 1.0112002200912684e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 393000, "loss": 0.5575108397752047, "lr": 0.0003, "grad_norm": 0.14547461456295072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 57524412536, "tokens_per_second": 148835.09593212197, "data_wait_s": 8.709001122042537e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1cef271e249b461e2526e3a9b9641cb73f40762214677051e73722e0994fa092
-size 808368
+oid sha256:20c89420c57af9ad9c54d1c03536403893acc59d116cc6d45220101ececa5890
+size 810486