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": 75314.93939064362,
-      "required_pages_per_second_upper_estimate": 0.006528108062412249
+      "required_token_positions_per_second": 73393.9785834345,
+      "required_pages_per_second_upper_estimate": 0.0063616040482740175
     },
     {
       "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": 9414.367423830452,
-      "required_pages_per_second_upper_estimate": 0.0037555319227024304
+      "required_token_positions_per_second": 9174.247322929312,
+      "required_pages_per_second_upper_estimate": 0.003659744424337527
     },
     {
       "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": 18828.734847660904,
-      "required_pages_per_second_upper_estimate": 0.004355696432590873
+      "required_token_positions_per_second": 18348.494645858624,
+      "required_pages_per_second_upper_estimate": 0.004244601313842798
     },
     {
       "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": 28243.102271491356,
-      "required_pages_per_second_upper_estimate": 0.001719856748228167
+      "required_token_positions_per_second": 27522.741968787937,
+      "required_pages_per_second_upper_estimate": 0.0016759905852319467
     },
     {
       "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": 28243.102271491356,
-      "required_pages_per_second_upper_estimate": 0.002174232194770357
+      "required_token_positions_per_second": 27522.741968787937,
+      "required_pages_per_second_upper_estimate": 0.0021187768645833027
     },
     {
       "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": 28243.102271491356,
-      "required_pages_per_second_upper_estimate": 0.3506848066291438
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27522.741968787937,
+      "required_pages_per_second_upper_estimate": 0.34174034256041247
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.12147670387554399,
-    "producer_p95_seconds": 0.3916274030052591,
-    "gpu_update_seconds": 0.9789292880870224,
-    "producer_batches_per_second": 8.232031065186998,
-    "gpu_batches_per_second": 1.0215242430371585,
-    "producer_headroom": 8.05857631015376,
+    "producer_batch_seconds": 0.1482706625934611,
+    "producer_p95_seconds": 0.3939153309984249,
+    "gpu_update_seconds": 1.00455107384846,
+    "producer_batches_per_second": 6.744422547984898,
+    "gpu_batches_per_second": 0.9954695445886841,
+    "producer_headroom": 6.775116913065995,
     "sustainable": true,
-    "buffer_seconds": 60.693615861395386,
+    "buffer_seconds": 60.273064430907596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414912.35746256396,
-    "upload_bytes_per_second": 3868512.749272909,
+    "generation_bytes_per_second": 430340.3663748679,
+    "upload_bytes_per_second": 3837045.406448443,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.03053324611368,
+    "estimated_full_shards_per_hour": 92.34102481302764,
     "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": 1.1924678350005706,
-      "bytes": 7370627,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 0.999219997000182,
       "bytes": 6670501,
@@ -562,6 +556,12 @@
       "bytes": 11951922,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8686913920028019,
+      "bytes": 2634908,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -356,3 +356,4 @@
 {"step": 350600, "loss": 0.588532218709588, "lr": 0.0003, "grad_norm": 0.16141536980723187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51318537286, "tokens_per_second": 148859.55574717984, "data_wait_s": 1.3267999747768044e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350700, "loss": 0.5680049220100045, "lr": 0.0003, "grad_norm": 0.15940457579135126, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51333187567, "tokens_per_second": 149293.90727806013, "data_wait_s": 1.622399577172473e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 350800, "loss": 0.5778562966734171, "lr": 0.0003, "grad_norm": 0.16013492337526053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51347915416, "tokens_per_second": 148592.92630792255, "data_wait_s": 1.1156000255141407e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 350900, "loss": 0.610313281416893, "lr": 0.0003, "grad_norm": 0.15464998559750495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51362586949, "tokens_per_second": 149301.19764136526, "data_wait_s": 8.33100057207048e-06, "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 sha256:036391e48e4e8fc29826a12087336050f7a3a90846844944aa876810b8f9cd8d
-size 763890
+oid sha256:01a3a17e9c1d9e64de01d8a8f310e89be54d4e7009c57d56b47313452fdd6298
+size 766008