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": 74910.67701832732,
-      "required_pages_per_second_upper_estimate": 0.0064930676245734875
+      "required_token_positions_per_second": 75355.77308669794,
+      "required_pages_per_second_upper_estimate": 0.006531647423694181
     },
     {
       "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": 9363.834627290915,
-      "required_pages_per_second_upper_estimate": 0.003735373634630172
+      "required_token_positions_per_second": 9419.471635837243,
+      "required_pages_per_second_upper_estimate": 0.003757568069186709
     },
     {
       "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": 18727.66925458183,
-      "required_pages_per_second_upper_estimate": 0.004332316686325718
+      "required_token_positions_per_second": 18838.943271674485,
+      "required_pages_per_second_upper_estimate": 0.004358057971824289
     },
     {
       "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": 28091.503881872744,
-      "required_pages_per_second_upper_estimate": 0.0017106252016756741
+      "required_token_positions_per_second": 28258.414907511728,
+      "required_pages_per_second_upper_estimate": 0.0017207892074226155
     },
     {
       "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": 28091.503881872744,
-      "required_pages_per_second_upper_estimate": 0.002162561731086315
+      "required_token_positions_per_second": 28258.414907511728,
+      "required_pages_per_second_upper_estimate": 0.002175411003879295
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28091.503881872744,
-      "required_pages_per_second_upper_estimate": 0.34880246199725273
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28258.414907511728,
+      "required_pages_per_second_upper_estimate": 0.35087493832042077
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10712037794530715,
-    "producer_p95_seconds": 0.34230192599716247,
-    "gpu_update_seconds": 0.9842121702085542,
-    "producer_batches_per_second": 9.335291932134274,
-    "gpu_batches_per_second": 1.0160410836904206,
-    "producer_headroom": 9.18790793205628,
+    "producer_batch_seconds": 0.21612502935144562,
+    "producer_p95_seconds": 0.34198722200017073,
+    "gpu_update_seconds": 0.9783988270570171,
+    "producer_batches_per_second": 4.626951366997287,
+    "gpu_batches_per_second": 1.0220780854858118,
+    "producer_headroom": 4.527003790320007,
     "sustainable": true,
-    "buffer_seconds": 60.036942382721804,
+    "buffer_seconds": 60.66072727753506,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 467843.5584049383,
-    "upload_bytes_per_second": 4094686.3531147796,
+    "generation_bytes_per_second": 465182.8805130567,
+    "upload_bytes_per_second": 4179249.6480450425,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.38833679305183,
+    "estimated_full_shards_per_hour": 99.81741725486542,
     "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": 0.9841479640017496,
-      "bytes": 2601592,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1784665140003199,
       "bytes": 6252756,
@@ -562,6 +556,12 @@
       "bytes": 7497853,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.1582007809993229,
+      "bytes": 12016223,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -236,3 +236,4 @@
 {"step": 108600, "loss": 0.604346277192235, "lr": 0.0003, "grad_norm": 0.14303170411994448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15891525165, "tokens_per_second": 147561.17176263384, "data_wait_s": 1.471900031901896e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108700, "loss": 0.5989245846867561, "lr": 0.0003, "grad_norm": 0.1259789825213902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15906098309, "tokens_per_second": 147376.48087369718, "data_wait_s": 1.2828997569158673e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 108800, "loss": 0.6310719847679138, "lr": 0.0003, "grad_norm": 0.1289245080110836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15920844965, "tokens_per_second": 149894.3880822485, "data_wait_s": 1.9346000044606626e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 108900, "loss": 0.598317040130496, "lr": 0.0003, "grad_norm": 0.13598713407394408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15935566462, "tokens_per_second": 149171.47226963693, "data_wait_s": 1.5513000107603148e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c8a89db73a217325600129450aa3ee2912e19b71eb4d79f67b7b77e46a002978
-size 507381
+oid sha256:2d8df95e78706515868a040c4edabf3790b2e8157b7a6570d169f8637e520996
+size 509499