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": 75047.3709570562,
-      "required_pages_per_second_upper_estimate": 0.006504915908734888
+      "required_token_positions_per_second": 74828.03381926876,
+      "required_pages_per_second_upper_estimate": 0.006485904321536365
     },
     {
       "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": 9380.921369632026,
-      "required_pages_per_second_upper_estimate": 0.003742189791619605
+      "required_token_positions_per_second": 9353.504227408595,
+      "required_pages_per_second_upper_estimate": 0.0037312526836638725
     },
     {
       "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": 18761.84273926405,
-      "required_pages_per_second_upper_estimate": 0.00434022212057416
+      "required_token_positions_per_second": 18707.00845481719,
+      "required_pages_per_second_upper_estimate": 0.004327537173917825
     },
     {
       "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": 28142.76410889608,
-      "required_pages_per_second_upper_estimate": 0.001713746680559769
+      "required_token_positions_per_second": 28060.512682225788,
+      "required_pages_per_second_upper_estimate": 0.0017087380002154327
     },
     {
       "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": 28142.76410889608,
-      "required_pages_per_second_upper_estimate": 0.0021665078852599615
+      "required_token_positions_per_second": 28060.512682225788,
+      "required_pages_per_second_upper_estimate": 0.002160175942748361
     },
     {
       "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": 28142.76410889608,
-      "required_pages_per_second_upper_estimate": 0.3494389424599387
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28060.512682225788,
+      "required_pages_per_second_upper_estimate": 0.3484176550185106
     }
   ],
   "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": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23599696233600298,
-    "producer_p95_seconds": 0.3152759029981098,
-    "gpu_update_seconds": 0.9824194913128778,
-    "producer_batches_per_second": 4.237342676369877,
-    "gpu_batches_per_second": 1.017895113892364,
-    "producer_headroom": 4.162848036637643,
+    "producer_batch_seconds": 0.09439860396102517,
+    "producer_p95_seconds": 0.32911561600121786,
+    "gpu_update_seconds": 0.9852991751470356,
+    "producer_batches_per_second": 10.593376999652188,
+    "gpu_batches_per_second": 1.0149201635643008,
+    "producer_headroom": 10.437645619778879,
     "sustainable": true,
-    "buffer_seconds": 60.91000846139842,
+    "buffer_seconds": 60.10324968396917,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 470155.8455713172,
-    "upload_bytes_per_second": 4033257.4730252796,
+    "generation_bytes_per_second": 488633.5744452333,
+    "upload_bytes_per_second": 4102653.020833448,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.88449979166639,
+    "estimated_full_shards_per_hour": 104.84939026849507,
     "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": 0.8785766789987974,
-      "bytes": 2610651,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0569672790006734,
       "bytes": 5933537,
@@ -562,6 +556,12 @@
       "bytes": 1087871,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.300397458999214,
+      "bytes": 11506344,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -197,3 +197,4 @@
 {"step": 104700, "loss": 0.5540802646428347, "lr": 0.0003, "grad_norm": 0.13967225520373433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15321526734, "tokens_per_second": 148942.31345991525, "data_wait_s": 1.6287001926684752e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 104800, "loss": 0.610983818769455, "lr": 0.0003, "grad_norm": 0.13781168663160606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15336195977, "tokens_per_second": 148995.1361121154, "data_wait_s": 1.850099943112582e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 104900, "loss": 0.5931651797145605, "lr": 0.0003, "grad_norm": 0.12879474546570285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15350920625, "tokens_per_second": 148624.37984174673, "data_wait_s": 1.4916997315594926e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 105000, "loss": 0.6216882299631834, "lr": 0.0003, "grad_norm": 0.13170376331896053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 15365640815, "tokens_per_second": 148940.21746605175, "data_wait_s": 1.3888999092159793e-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:a64dbf6b5ade5d76478308cea24f91170fa7508e61c3de77b7909607232f5379
-size 423996
+oid sha256:dafb5ee794ede0bd3e62224bfd644c0b8c41e3c90fd69e29733e8c3b251fec7e
+size 426114