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": 75676.27091991807,
-      "required_pages_per_second_upper_estimate": 0.0065594273635833705
+      "required_token_positions_per_second": 74336.64555505331,
+      "required_pages_per_second_upper_estimate": 0.00644331203220634
     },
     {
       "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": 9459.533864989759,
-      "required_pages_per_second_upper_estimate": 0.0037735494913793514
+      "required_token_positions_per_second": 9292.080694381664,
+      "required_pages_per_second_upper_estimate": 0.0037067499179757714
     },
     {
       "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": 18919.067729979517,
-      "required_pages_per_second_upper_estimate": 0.004376593355110854
+      "required_token_positions_per_second": 18584.16138876333,
+      "required_pages_per_second_upper_estimate": 0.004299118667220784
     },
     {
       "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": 28378.601594969277,
-      "required_pages_per_second_upper_estimate": 0.0017281079461179641
+      "required_token_positions_per_second": 27876.242083144993,
+      "required_pages_per_second_upper_estimate": 0.0016975168875245279
     },
     {
       "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": 28378.601594969277,
-      "required_pages_per_second_upper_estimate": 0.0021846633077778213
+      "required_token_positions_per_second": 27876.242083144993,
+      "required_pages_per_second_upper_estimate": 0.0021459902819374556
     },
     {
       "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": 28378.601594969277,
-      "required_pages_per_second_upper_estimate": 0.3523672547396759
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27876.242083144993,
+      "required_pages_per_second_upper_estimate": 0.3461296308919502
     }
   ],
   "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": 0.3373416699608356,
-    "producer_p95_seconds": 0.3372497860000294,
-    "gpu_update_seconds": 0.9742551939169972,
-    "producer_batches_per_second": 2.9643536184429786,
-    "gpu_batches_per_second": 1.0264251155587847,
-    "producer_headroom": 2.8880369093747165,
+    "producer_batch_seconds": 0.2406260255860957,
+    "producer_p95_seconds": 0.32399118700050167,
+    "gpu_update_seconds": 0.9918123080412264,
+    "producer_batches_per_second": 4.155826442980505,
+    "gpu_batches_per_second": 1.0082552836785659,
+    "producer_headroom": 4.121799816231254,
     "sustainable": true,
-    "buffer_seconds": 60.40382202285383,
+    "buffer_seconds": 60.500550790514815,
     "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": 476748.6250250998,
-    "upload_bytes_per_second": 3959372.8376491196,
+    "generation_bytes_per_second": 453462.4581176321,
+    "upload_bytes_per_second": 4036498.9242461696,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 102.29915679039712,
+    "estimated_full_shards_per_hour": 97.30248744627687,
     "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": 1.1755309580003086,
-      "bytes": 7041796,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 0.8336417800001072,
       "bytes": 5838305,
@@ -562,6 +556,12 @@
       "bytes": 1174580,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.8239342810011294,
+      "bytes": 17807257,
+      "pages": 31,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -215,3 +215,4 @@
 {"step": 136500, "loss": 0.548358827829361, "lr": 0.0003, "grad_norm": 0.15211027111072187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19976590796, "tokens_per_second": 148067.31253867495, "data_wait_s": 1.1274998541921377e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 136600, "loss": 0.617629662156105, "lr": 0.0003, "grad_norm": 0.15347404341235923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 19991072917, "tokens_per_second": 147470.402349882, "data_wait_s": 1.2537999282358214e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 136700, "loss": 0.6210627965629101, "lr": 0.0003, "grad_norm": 0.15209265436022484, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20005535906, "tokens_per_second": 147356.34552390876, "data_wait_s": 1.413200152455829e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 136800, "loss": 0.5940307341516018, "lr": 0.0003, "grad_norm": 0.15701820478888792, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 20020180662, "tokens_per_second": 148697.8069655777, "data_wait_s": 6.69799919705838e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:57aa06ce6739f1d97def89320a3a559e9937adbf7f3515466c29f3a30349a12b
-size 462903
+oid sha256:1fcc034b7bac5fe56bde1bc86ffd0850c6ca97745f4b01d796176f217ab7b03a
+size 465021