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": 75023.60866319496,
-      "required_pages_per_second_upper_estimate": 0.006502856253327978
+      "required_token_positions_per_second": 74079.1455526633,
+      "required_pages_per_second_upper_estimate": 0.0064209925846269635
     },
     {
       "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": 9377.95108289937,
-      "required_pages_per_second_upper_estimate": 0.0037410048998322044
+      "required_token_positions_per_second": 9259.893194082913,
+      "required_pages_per_second_upper_estimate": 0.0036939098428605847
     },
     {
       "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": 18755.90216579874,
-      "required_pages_per_second_upper_estimate": 0.004338847873453483
+      "required_token_positions_per_second": 18519.786388165827,
+      "required_pages_per_second_upper_estimate": 0.004284226643793867
     },
     {
       "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": 28133.853248698106,
-      "required_pages_per_second_upper_estimate": 0.0017132040559253827
+      "required_token_positions_per_second": 27779.679582248744,
+      "required_pages_per_second_upper_estimate": 0.0016916367378442345
     },
     {
       "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": 28133.853248698106,
-      "required_pages_per_second_upper_estimate": 0.0021658219025750816
+      "required_token_positions_per_second": 27779.679582248744,
+      "required_pages_per_second_upper_estimate": 0.002138556633316352
     },
     {
       "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": 28133.853248698106,
-      "required_pages_per_second_upper_estimate": 0.34932829939901044
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27779.679582248744,
+      "required_pages_per_second_upper_estimate": 0.3449306478047201
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.134094598320587,
-    "producer_p95_seconds": 0.3629163799996604,
-    "gpu_update_seconds": 0.9827306539064076,
-    "producer_batches_per_second": 7.457421943345156,
-    "gpu_batches_per_second": 1.0175728171548795,
-    "producer_headroom": 7.328637142839577,
+    "producer_batch_seconds": 0.1043351397188701,
+    "producer_p95_seconds": 0.30919408800036763,
+    "gpu_update_seconds": 0.9952598595725746,
+    "producer_batches_per_second": 9.584498594572157,
+    "gpu_batches_per_second": 1.004762716371844,
+    "producer_headroom": 9.539066725307423,
     "sustainable": true,
-    "buffer_seconds": 60.92930054219727,
+    "buffer_seconds": 60.71085143392705,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 408098.7607021228,
-    "upload_bytes_per_second": 2924251.080906278,
+    "generation_bytes_per_second": 424620.5131596053,
+    "upload_bytes_per_second": 3012159.157027058,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.56849399373783,
+    "estimated_full_shards_per_hour": 91.11367746440048,
     "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": 6.862793137002882,
-      "bytes": 7606529,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 3.3253811320028035,
       "bytes": 1169016,
@@ -562,6 +556,12 @@
       "bytes": 9747301,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.9869726110009651,
+      "bytes": 2644359,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -227,3 +227,4 @@
 {"step": 302900, "loss": 0.6034968458116055, "lr": 0.0003, "grad_norm": 0.1457596251327611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44336969250, "tokens_per_second": 150457.0899802314, "data_wait_s": 1.020499985315837e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303000, "loss": 0.5826489478349686, "lr": 0.0003, "grad_norm": 0.15378741494797135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44351668829, "tokens_per_second": 149681.03496009723, "data_wait_s": 1.1145002645207569e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 303100, "loss": 0.5596464416012168, "lr": 0.0003, "grad_norm": 0.1396069368276492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44366233356, "tokens_per_second": 149082.00414015295, "data_wait_s": 1.4781999198021367e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 303200, "loss": 0.7135458812117577, "lr": 0.0003, "grad_norm": 0.22248423776876727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44380786988, "tokens_per_second": 148440.79297462487, "data_wait_s": 1.4043998817214742e-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:f49dcb7de46509480119f76919f8a161ab343ce04d67c36803d89437692f4c93
-size 488319
+oid sha256:c25f1e5365dbac4cfe070bd5e00de02cc4d6d6ba2959c09d48e73da677e4124b
+size 490437