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": 74246.27787194915,
-      "required_pages_per_second_upper_estimate": 0.00643547919046967
+      "required_token_positions_per_second": 75229.80294535331,
+      "required_pages_per_second_upper_estimate": 0.006520728651110834
     },
     {
       "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": 9280.784733993643,
-      "required_pages_per_second_upper_estimate": 0.003702243790487332
+      "required_token_positions_per_second": 9403.725368169164,
+      "required_pages_per_second_upper_estimate": 0.003751286647586231
     },
     {
       "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": 18561.569467987287,
-      "required_pages_per_second_upper_estimate": 0.004293892423953473
+      "required_token_positions_per_second": 18807.450736338327,
+      "required_pages_per_second_upper_estimate": 0.004350772728023961
     },
     {
       "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": 27842.354201980932,
-      "required_pages_per_second_upper_estimate": 0.0016954532933576084
+      "required_token_positions_per_second": 28211.17610450749,
+      "required_pages_per_second_upper_estimate": 0.0017179126121625024
     },
     {
       "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": 27842.354201980932,
-      "required_pages_per_second_upper_estimate": 0.002143381499037787
+      "required_token_positions_per_second": 28211.17610450749,
+      "required_pages_per_second_upper_estimate": 0.0021717744300586523
     },
     {
       "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": 27842.354201980932,
-      "required_pages_per_second_upper_estimate": 0.345708856823335
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28211.17610450749,
+      "required_pages_per_second_upper_estimate": 0.35028839048521165
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 690670,
-    "producer_batch_seconds": 0.09368471524218336,
-    "producer_p95_seconds": 0.1963417759998265,
-    "gpu_update_seconds": 0.9930194767090814,
-    "producer_batches_per_second": 10.674099797548733,
-    "gpu_batches_per_second": 1.0070295935322964,
-    "producer_headroom": 10.599588995302353,
+    "producer_batch_seconds": 0.05004556503124746,
+    "producer_p95_seconds": 0.18120563000047696,
+    "gpu_update_seconds": 0.9800371277531564,
+    "producer_batches_per_second": 19.98179058175524,
+    "gpu_batches_per_second": 1.0203695060947442,
+    "producer_headroom": 19.58289664910848,
     "sustainable": true,
-    "buffer_seconds": 60.574188079253965,
+    "buffer_seconds": 60.7623019206957,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432844.96831240057,
+    "generation_bytes_per_second": 433279.75571630104,
     "upload_bytes_per_second": 2970521.1586212316,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.87845408467305,
+    "estimated_full_shards_per_hour": 92.97174934021733,
     "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": [

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -45,3 +45,4 @@
 {"step": 241600, "loss": 0.6268334165215492, "lr": 0.0003, "grad_norm": 0.14220717144270645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35363108475, "tokens_per_second": 148962.41087267562, "data_wait_s": 9.933000001183245e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 241700, "loss": 0.5771536789834499, "lr": 0.0003, "grad_norm": 0.14329444178759554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35377628896, "tokens_per_second": 148390.99048776756, "data_wait_s": 9.821000276133418e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 241800, "loss": 0.5467068497091532, "lr": 0.0003, "grad_norm": 0.1420090650479915, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35392095639, "tokens_per_second": 146137.013868926, "data_wait_s": 1.1516999620653223e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 241900, "loss": 0.6098959427326918, "lr": 0.0003, "grad_norm": 0.14584189645100581, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35406558433, "tokens_per_second": 147494.01958781225, "data_wait_s": 1.374099974782439e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:edd325d48385e7037d6f04b4f0bb87e03d0638bc11d5168c2e2985b29b966063
-size 100494
+oid sha256:28fdf4ef4f9e53e392fa6fbf944759459f5d4a807babfe2f3ac28fdbf05ac47c
+size 102612