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": 73459.22885286577,
-      "required_pages_per_second_upper_estimate": 0.006367259776253024
+      "required_token_positions_per_second": 75861.03871064603,
+      "required_pages_per_second_upper_estimate": 0.006575442567393985
     },
     {
       "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": 9182.40360660822,
-      "required_pages_per_second_upper_estimate": 0.003662998087844352
+      "required_token_positions_per_second": 9482.629838830753,
+      "required_pages_per_second_upper_estimate": 0.0037827628206601057
     },
     {
       "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": 18364.80721321644,
-      "required_pages_per_second_upper_estimate": 0.0042483749392096485
+      "required_token_positions_per_second": 18965.259677661506,
+      "required_pages_per_second_upper_estimate": 0.0043872790492565684
     },
     {
       "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": 27547.21081982466,
-      "required_pages_per_second_upper_estimate": 0.0016774806098819434
+      "required_token_positions_per_second": 28447.889516492258,
+      "required_pages_per_second_upper_estimate": 0.0017323272170130842
     },
     {
       "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": 27547.21081982466,
-      "required_pages_per_second_upper_estimate": 0.0021206605444702245
+      "required_token_positions_per_second": 28447.889516492258,
+      "required_pages_per_second_upper_estimate": 0.002189997283777903
     },
     {
       "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": 27547.21081982466,
-      "required_pages_per_second_upper_estimate": 0.3420441637983121
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28447.889516492258,
+      "required_pages_per_second_upper_estimate": 0.35322757883323513
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.4697109957578789,
-    "producer_p95_seconds": 0.25972810500024934,
-    "gpu_update_seconds": 1.003658779861038,
-    "producer_batches_per_second": 2.128968682937685,
-    "gpu_batches_per_second": 0.9963545580087044,
-    "producer_headroom": 2.136758110679598,
+    "producer_batch_seconds": 0.11328751029682849,
+    "producer_p95_seconds": 0.2701105159976578,
+    "gpu_update_seconds": 0.9718822896851967,
+    "producer_batches_per_second": 8.827098392222283,
+    "gpu_batches_per_second": 1.0289311891092396,
+    "producer_headroom": 8.578900596709511,
     "sustainable": true,
-    "buffer_seconds": 60.219526791662275,
+    "buffer_seconds": 60.2567019604822,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288679.67688002595,
+    "generation_bytes_per_second": 301101.7482841268,
     "upload_bytes_per_second": 5129014.020471953,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.94393853951057,
+    "estimated_full_shards_per_hour": 64.60942589180806,
     "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

@@ -180,3 +180,4 @@
 {"step": 255100, "loss": 0.6156003326177597, "lr": 0.0003, "grad_norm": 0.14337446662061135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37339925108, "tokens_per_second": 132571.0288182981, "data_wait_s": 1.1971998901572078e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255200, "loss": 0.592559328302741, "lr": 0.0003, "grad_norm": 0.1478354914599657, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37354605934, "tokens_per_second": 149664.23562165003, "data_wait_s": 9.801999112823978e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 255300, "loss": 0.6864411178976297, "lr": 0.0003, "grad_norm": 0.17404576282657638, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37369247230, "tokens_per_second": 149035.1470873743, "data_wait_s": 6.710000889142975e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 255400, "loss": 0.5846588611602783, "lr": 0.0003, "grad_norm": 0.13649780259987665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37383657652, "tokens_per_second": 146384.4460172169, "data_wait_s": 1.2704000255325809e-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 sha256:86f87b416ef84fc4134ae13fc65a2e03b00849604ff138d2042f4740d7a0b551
-size 388773
+oid sha256:81d58be49aa0990ad2296e3f33d2c240d12e7bfc622c64e19dccabd221b68cb5
+size 390891