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": 75761.77729758614,
-      "required_pages_per_second_upper_estimate": 0.006566838839685707
+      "required_token_positions_per_second": 76317.68258985099,
+      "required_pages_per_second_upper_estimate": 0.00661502330149014
     },
     {
       "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": 9470.222162198268,
-      "required_pages_per_second_upper_estimate": 0.00377781321294011
+      "required_token_positions_per_second": 9539.710323731373,
+      "required_pages_per_second_upper_estimate": 0.003805533079516265
     },
     {
       "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": 18940.444324396536,
-      "required_pages_per_second_upper_estimate": 0.004381538454014019
+      "required_token_positions_per_second": 19079.420647462746,
+      "required_pages_per_second_upper_estimate": 0.004413688180455638
     },
     {
       "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": 28410.666486594804,
-      "required_pages_per_second_upper_estimate": 0.001730060529786474
+      "required_token_positions_per_second": 28619.130971194118,
+      "required_pages_per_second_upper_estimate": 0.0017427549231699493
     },
     {
       "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": 28410.666486594804,
-      "required_pages_per_second_upper_estimate": 0.0021871317518964554
+      "required_token_positions_per_second": 28619.130971194118,
+      "required_pages_per_second_upper_estimate": 0.002203179924987538
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28410.666486594804,
-      "required_pages_per_second_upper_estimate": 0.352765393379376
+      "required_token_positions_per_second": 28619.130971194118,
+      "required_pages_per_second_upper_estimate": 0.3553538245923503
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.280893026031265,
-    "producer_p95_seconds": 0.3834357030063984,
-    "gpu_update_seconds": 0.9731556284695166,
-    "producer_batches_per_second": 3.560074146834441,
-    "gpu_batches_per_second": 1.027584870030194,
-    "producer_headroom": 3.4645061937607484,
+    "producer_batch_seconds": 0.10905964226509468,
+    "producer_p95_seconds": 0.3740103449963499,
+    "gpu_update_seconds": 0.9660670698851204,
+    "producer_batches_per_second": 9.169294701785915,
+    "gpu_batches_per_second": 1.035124818113213,
+    "producer_headroom": 8.858153665467478,
     "sustainable": true,
-    "buffer_seconds": 60.33564896511003,
+    "buffer_seconds": 60.86222540276258,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456378.314368499,
-    "upload_bytes_per_second": 5625832.0593866585,
+    "generation_bytes_per_second": 456978.8098146296,
+    "upload_bytes_per_second": 5725380.974174499,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.92816232005336,
+    "estimated_full_shards_per_hour": 98.05701466397444,
     "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": 1.3357893329994113,
-      "bytes": 5825933,
-      "pages": 29,
-      "shards": 2
-    },
     {
       "seconds": 1.0486931550003646,
       "bytes": 6833428,
@@ -562,6 +556,12 @@
       "bytes": 6809793,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.123792185993807,
+      "bytes": 11983615,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -305,3 +305,4 @@
 {"step": 310700, "loss": 0.5604292182251811, "lr": 0.0003, "grad_norm": 0.1467913101522624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45478207863, "tokens_per_second": 149486.1072726009, "data_wait_s": 1.9542996597010642e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 310800, "loss": 0.6214486751705408, "lr": 0.0003, "grad_norm": 0.1710393720838941, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45492806850, "tokens_per_second": 149143.65493830875, "data_wait_s": 9.006995242089033e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 310900, "loss": 0.5864830911159515, "lr": 0.0003, "grad_norm": 0.1517254468936491, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45507580553, "tokens_per_second": 151239.32859975062, "data_wait_s": 8.03900184109807e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 311000, "loss": 0.5390106067061424, "lr": 0.0003, "grad_norm": 0.13726870435127175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 45522397256, "tokens_per_second": 152254.52151105856, "data_wait_s": 1.030000566970557e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6e88c2df1b0481ec5f7fca96b9ab59091a9a0591a001f5904de4ab29d4c127e2
-size 655089
+oid sha256:6fc4e04ccc8e43138e98cbce7307f4b160f2ea3d4ce8257089f3bed539f5339e
+size 657207