Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74631.28386319977,
-      "required_pages_per_second_upper_estimate": 0.005163165386092678
+      "required_token_positions_per_second": 76634.88887939186,
+      "required_pages_per_second_upper_estimate": 0.005301779430116991
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9328.910482899972,
-      "required_pages_per_second_upper_estimate": 0.003905820009386774
+      "required_token_positions_per_second": 9579.361109923982,
+      "required_pages_per_second_upper_estimate": 0.004010678456917912
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18657.820965799943,
-      "required_pages_per_second_upper_estimate": 0.0038859320673490147
+      "required_token_positions_per_second": 19158.722219847965,
+      "required_pages_per_second_upper_estimate": 0.003990256588912838
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27986.731448699917,
-      "required_pages_per_second_upper_estimate": 0.0016777618170849742
+      "required_token_positions_per_second": 28738.083329771947,
+      "required_pages_per_second_upper_estimate": 0.0017228042151073478
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27986.731448699917,
-      "required_pages_per_second_upper_estimate": 0.0021758452695144503
+      "required_token_positions_per_second": 28738.083329771947,
+      "required_pages_per_second_upper_estimate": 0.002234259573420143
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 27986.731448699917,
-      "required_pages_per_second_upper_estimate": 0.3589837412129131
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28738.083329771947,
+      "required_pages_per_second_upper_estimate": 0.36862127640450926
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.28727027044539,
-    "producer_p95_seconds": 0.3489155660063261,
-    "gpu_update_seconds": 0.9878967127933173,
-    "producer_batches_per_second": 3.4810424289627275,
-    "gpu_batches_per_second": 1.0122515714952227,
-    "producer_headroom": 3.438910372666343,
+    "producer_batch_seconds": 0.2704173131485845,
+    "producer_p95_seconds": 0.3490291159978369,
+    "gpu_update_seconds": 0.9620683356902008,
+    "producer_batches_per_second": 3.6979880775996623,
+    "gpu_batches_per_second": 1.0394272037677932,
+    "producer_headroom": 3.5577172352185125,
     "sustainable": true,
-    "buffer_seconds": 60.26169948039235,
+    "buffer_seconds": 60.61030514848265,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370015.52508916426,
-    "upload_bytes_per_second": 4837395.222574953,
+    "generation_bytes_per_second": 355499.84227185056,
+    "upload_bytes_per_second": 4814130.459829377,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.39671816354938,
+    "estimated_full_shards_per_hour": 76.28199053875578,
     "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": 1.4925930749996041,
-      "bytes": 12021057,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.3497725520028325,
       "bytes": 6087212,
@@ -562,6 +556,12 @@
       "bytes": 2348648,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.589205835996836,
+      "bytes": 10428035,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -362,3 +362,4 @@
 {"step": 640600, "loss": 0.5625096559524536, "lr": 0.0001, "grad_norm": 0.19436205433084347, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93767531675, "tokens_per_second": 153343.26154659578, "data_wait_s": 1.4709003153257072e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 640700, "loss": 0.5348134636878967, "lr": 0.0001, "grad_norm": 0.1798952781332512, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93782216773, "tokens_per_second": 153296.59126634218, "data_wait_s": 1.3729004422202706e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 640800, "loss": 0.5738946199417114, "lr": 0.0001, "grad_norm": 0.19433422428929165, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93796718183, "tokens_per_second": 150008.37083661344, "data_wait_s": 1.7571001080796123e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 640900, "loss": 0.5774219036102295, "lr": 0.0001, "grad_norm": 0.1634302727952373, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93811256697, "tokens_per_second": 151218.83768895429, "data_wait_s": 1.7651997040957212e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a3b7fb5af5fa59aba6b17a07db704075676595d88dcac2d9da9fa947b26d086a
-size 777381
+oid sha256:12f70c0c7f973999815c5c32e6246541fdc5f29b8c21f998ac61b567d99b1a37
+size 779499