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": 75001.79153109554,
-      "required_pages_per_second_upper_estimate": 0.006500965199612616
+      "required_token_positions_per_second": 74815.22412959313,
+      "required_pages_per_second_upper_estimate": 0.006484794010101143
     },
     {
       "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": 9375.223941386943,
-      "required_pages_per_second_upper_estimate": 0.003739917002308498
+      "required_token_positions_per_second": 9351.903016199141,
+      "required_pages_per_second_upper_estimate": 0.0037306139365721803
     },
     {
       "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": 18750.447882773886,
-      "required_pages_per_second_upper_estimate": 0.004337586121067785
+      "required_token_positions_per_second": 18703.806032398283,
+      "required_pages_per_second_upper_estimate": 0.00432679634985192
     },
     {
       "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": 28125.671824160832,
-      "required_pages_per_second_upper_estimate": 0.0017127058500956797
+      "required_token_positions_per_second": 28055.709048597426,
+      "required_pages_per_second_upper_estimate": 0.0017084454841301833
     },
     {
       "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": 28125.671824160832,
-      "required_pages_per_second_upper_estimate": 0.00216519207386657
+      "required_token_positions_per_second": 28055.709048597426,
+      "required_pages_per_second_upper_estimate": 0.0021598061457343406
     },
     {
       "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": 28125.671824160832,
-      "required_pages_per_second_upper_estimate": 0.34922671348772405
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28055.709048597426,
+      "required_pages_per_second_upper_estimate": 0.3483580099655739
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07293702549210224,
-    "producer_p95_seconds": 0.1942122760001439,
-    "gpu_update_seconds": 0.9830165186045264,
-    "producer_batches_per_second": 13.710457661976934,
-    "gpu_batches_per_second": 1.0172769033622986,
-    "producer_headroom": 13.47760635935132,
+    "producer_batch_seconds": 0.0885677412031356,
+    "producer_p95_seconds": 0.20553095800005394,
+    "gpu_update_seconds": 0.9854678757934365,
+    "producer_batches_per_second": 11.290792634153762,
+    "gpu_batches_per_second": 1.0147464210285526,
+    "producer_headroom": 11.126713433203687,
     "sustainable": true,
-    "buffer_seconds": 60.94702415348064,
+    "buffer_seconds": 60.11354042339963,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369119.71125756303,
+    "generation_bytes_per_second": 369786.89977421705,
     "upload_bytes_per_second": 6020291.987130927,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.20449736876648,
+    "estimated_full_shards_per_hour": 79.34766049308666,
     "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-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -74,3 +74,4 @@
 {"step": 89300, "loss": 0.6066636741161346, "lr": 0.0003, "grad_norm": 0.12688301071404934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13068424989, "tokens_per_second": 148636.4784598391, "data_wait_s": 1.2800000149582047e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 89400, "loss": 0.5993508864194155, "lr": 0.0003, "grad_norm": 0.14219732352287673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13083191395, "tokens_per_second": 150169.57744113944, "data_wait_s": 8.537000212527346e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 89500, "loss": 0.6130473129451275, "lr": 0.0003, "grad_norm": 0.14434399898543035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13097858409, "tokens_per_second": 148655.6382351637, "data_wait_s": 1.8044000171357766e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 89600, "loss": 0.5701491050422192, "lr": 0.0003, "grad_norm": 0.13772993576607845, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13112480965, "tokens_per_second": 142778.85291466417, "data_wait_s": 1.617900124983862e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f456dbbcc4cfdbec8f2fefff86c5a2a39bcf9065b9ba52621703188c7ddfd2f8
-size 161916
+oid sha256:05358c7d81e13d8dc5aec60406e6805fd602bf3992719353c92ae51094a96ae4
+size 164034