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": 75396.39935795023,
-      "required_pages_per_second_upper_estimate": 0.006535168805919986
+      "required_token_positions_per_second": 75227.74145594588,
+      "required_pages_per_second_upper_estimate": 0.006520549966433795
     },
     {
       "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": 9424.549919743778,
-      "required_pages_per_second_upper_estimate": 0.003759593872564137
+      "required_token_positions_per_second": 9403.467681993236,
+      "required_pages_per_second_upper_estimate": 0.0037511838527179014
     },
     {
       "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": 18849.099839487557,
-      "required_pages_per_second_upper_estimate": 0.004360407515038354
+      "required_token_positions_per_second": 18806.93536398647,
+      "required_pages_per_second_upper_estimate": 0.004350653505700602
     },
     {
       "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": 28273.649759231335,
-      "required_pages_per_second_upper_estimate": 0.00172171692996125
+      "required_token_positions_per_second": 28210.403045979707,
+      "required_pages_per_second_upper_estimate": 0.001717865536954085
     },
     {
       "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": 28273.649759231335,
-      "required_pages_per_second_upper_estimate": 0.002176583824937439
+      "required_token_positions_per_second": 28210.403045979707,
+      "required_pages_per_second_upper_estimate": 0.0021717149178732205
     },
     {
       "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": 28273.649759231335,
-      "required_pages_per_second_upper_estimate": 0.35106410419100953
+      "required_token_positions_per_second": 28210.403045979707,
+      "required_pages_per_second_upper_estimate": 0.35027879168555703
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.06690109020320278,
-    "producer_p95_seconds": 0.21632020699689747,
-    "gpu_update_seconds": 0.977871630845004,
-    "producer_batches_per_second": 14.947439525464214,
-    "gpu_batches_per_second": 1.0226291145555315,
-    "producer_headroom": 14.616677065722763,
+    "producer_batch_seconds": 0.4609511448828698,
+    "producer_p95_seconds": 0.2774087679972581,
+    "gpu_update_seconds": 0.9800639840181278,
+    "producer_batches_per_second": 2.1694273050435866,
+    "gpu_batches_per_second": 1.0203415453551687,
+    "producer_headroom": 2.1261775676187273,
     "sustainable": true,
-    "buffer_seconds": 60.62804111239024,
+    "buffer_seconds": 60.763967009123924,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294982.2025964519,
-    "upload_bytes_per_second": 5236085.920943062,
+    "generation_bytes_per_second": 289347.85124829283,
+    "upload_bytes_per_second": 5264266.185710132,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.29631384296577,
+    "estimated_full_shards_per_hour": 62.08731320463742,
     "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": 0.9283750220001821,
-      "bytes": 2657482,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9979908650002471,
       "bytes": 5962542,
@@ -562,6 +556,12 @@
       "bytes": 771997,
       "pages": 25,
       "shards": 1
+    },
+    {
+      "seconds": 1.993406634999701,
+      "bytes": 10476570,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -198,3 +198,4 @@
 {"step": 214800, "loss": 0.5876647941768169, "lr": 0.0003, "grad_norm": 0.14906130716057017, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31440150645, "tokens_per_second": 150202.25105826938, "data_wait_s": 1.3276003301143646e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 214900, "loss": 0.6138138696551323, "lr": 0.0003, "grad_norm": 0.15233410682666723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31454647483, "tokens_per_second": 147255.4126865786, "data_wait_s": 8.671999239595607e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 215000, "loss": 0.5686859451234341, "lr": 0.0003, "grad_norm": 0.1390351501732773, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31469240712, "tokens_per_second": 134576.64699822635, "data_wait_s": 1.1709002137649804e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 215100, "loss": 0.5917164199054241, "lr": 0.0003, "grad_norm": 0.1463686432349559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31483816673, "tokens_per_second": 128251.81969456565, "data_wait_s": 1.2594002328114584e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1b11531a059d9bcb204d29c20f2eea41759af548321c1e87c7d5ae3962faaca7
-size 426897
+oid sha256:59fa763dc1078fb88c2f5b6e64d8053e860c97e6c088021dd916708fad6dd502
+size 429015