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": 74551.7866533125,
-      "required_pages_per_second_upper_estimate": 0.006461959917333318
+      "required_token_positions_per_second": 74341.17926188819,
+      "required_pages_per_second_upper_estimate": 0.006443705002424212
     },
     {
       "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": 9318.973331664063,
-      "required_pages_per_second_upper_estimate": 0.003717477793068479
+      "required_token_positions_per_second": 9292.647407736024,
+      "required_pages_per_second_upper_estimate": 0.003706975988405945
     },
     {
       "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": 18637.946663328126,
-      "required_pages_per_second_upper_estimate": 0.004311560943902847
+      "required_token_positions_per_second": 18585.294815472047,
+      "required_pages_per_second_upper_estimate": 0.004299380865542217
     },
     {
       "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": 27956.91999499219,
-      "required_pages_per_second_upper_estimate": 0.0017024297490717318
+      "required_token_positions_per_second": 27877.94222320807,
+      "required_pages_per_second_upper_estimate": 0.0016976204171344237
     },
     {
       "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": 27956.91999499219,
-      "required_pages_per_second_upper_estimate": 0.0021522010909222067
+      "required_token_positions_per_second": 27877.94222320807,
+      "required_pages_per_second_upper_estimate": 0.0021461211634258
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27956.91999499219,
-      "required_pages_per_second_upper_estimate": 0.34713138054549075
+      "required_token_positions_per_second": 27877.94222320807,
+      "required_pages_per_second_upper_estimate": 0.3461507409415308
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11064839617975508,
-    "producer_p95_seconds": 0.33230805000130204,
-    "gpu_update_seconds": 0.9889501420382673,
-    "producer_batches_per_second": 9.03763664477738,
-    "gpu_batches_per_second": 1.0111733215781318,
-    "producer_headroom": 8.93777204354284,
+    "producer_batch_seconds": 0.09351600074981548,
+    "producer_p95_seconds": 0.2998225330011337,
+    "gpu_update_seconds": 0.9917518222339723,
+    "producer_batches_per_second": 10.693357200713837,
+    "gpu_batches_per_second": 1.0083167760130234,
+    "producer_headroom": 10.605156489606719,
     "sustainable": true,
-    "buffer_seconds": 60.3259586643343,
+    "buffer_seconds": 60.49686115627231,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 352077.3179464804,
-    "upload_bytes_per_second": 5541531.682535544,
+    "generation_bytes_per_second": 395380.14505130943,
+    "upload_bytes_per_second": 5460643.949395277,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.54759649082001,
+    "estimated_full_shards_per_hour": 84.83937514929258,
     "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": 1.0544361239990394,
-      "bytes": 7130727,
-      "pages": 44,
-      "shards": 2
-    },
     {
       "seconds": 1.1150904570004059,
       "bytes": 10599538,
@@ -562,6 +556,12 @@
       "bytes": 7398466,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.0474640769971302,
+      "bytes": 900238,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -214,3 +214,4 @@
 {"step": 336400, "loss": 0.5758977457880974, "lr": 0.0003, "grad_norm": 0.1447568800356445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49240325244, "tokens_per_second": 147742.40699916505, "data_wait_s": 1.1849999282276258e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 336500, "loss": 0.6084229610860348, "lr": 0.0003, "grad_norm": 0.16968612888862306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49254826301, "tokens_per_second": 147117.7818765244, "data_wait_s": 1.3072000001557171e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 336600, "loss": 0.5735634211450815, "lr": 0.0003, "grad_norm": 0.15155936902108486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49269462398, "tokens_per_second": 147894.26294284908, "data_wait_s": 1.3128999853506684e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 336700, "loss": 0.5935948863625526, "lr": 0.0003, "grad_norm": 0.14299183902306487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49284108645, "tokens_per_second": 148462.8509218014, "data_wait_s": 1.3539996871259063e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7e9a6dece0fe8da2857ecf4d2476445d6c13f1d1616f65b3cc05f81c0307ec9f
-size 460785
+oid sha256:073a6e1e951375fbdd9a796e49ec422cbbad739565a6839946090cd28bd0f0ff
+size 462903