Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77379.16614487718,
-      "required_pages_per_second_upper_estimate": 0.00535327025830445
+      "required_token_positions_per_second": 77435.04060553477,
+      "required_pages_per_second_upper_estimate": 0.0053571357831393536
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9672.395768109647,
-      "required_pages_per_second_upper_estimate": 0.004049630125515665
+      "required_token_positions_per_second": 9679.380075691846,
+      "required_pages_per_second_upper_estimate": 0.004052554309251405
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19344.791536219294,
-      "required_pages_per_second_upper_estimate": 0.0040290099205351326
+      "required_token_positions_per_second": 19358.760151383693,
+      "required_pages_per_second_upper_estimate": 0.004031919214696723
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29017.18730432894,
-      "required_pages_per_second_upper_estimate": 0.0017395360722149463
+      "required_token_positions_per_second": 29038.14022707554,
+      "required_pages_per_second_upper_estimate": 0.0017407921679403443
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29017.18730432894,
-      "required_pages_per_second_upper_estimate": 0.0022559586797933073
+      "required_token_positions_per_second": 29038.14022707554,
+      "required_pages_per_second_upper_estimate": 0.002257587677374688
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29017.18730432894,
-      "required_pages_per_second_upper_estimate": 0.37220132251162685
+      "required_token_positions_per_second": 29038.14022707554,
+      "required_pages_per_second_upper_estimate": 0.37247008410712457
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10757230974206777,
-    "producer_p95_seconds": 0.31675270199775696,
-    "gpu_update_seconds": 0.9528146098390219,
-    "producer_batches_per_second": 9.296072589663238,
-    "gpu_batches_per_second": 1.049522110254953,
-    "producer_headroom": 8.857433777555205,
+    "producer_batch_seconds": 0.1191641781175008,
+    "producer_p95_seconds": 0.29777948599803494,
+    "gpu_update_seconds": 0.9521270916041877,
+    "producer_batches_per_second": 8.39178363663918,
+    "gpu_batches_per_second": 1.0502799561297576,
+    "producer_headroom": 7.990044547324878,
     "sustainable": true,
-    "buffer_seconds": 60.02732041985838,
+    "buffer_seconds": 60.936133862668015,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 39431.68941601908,
-    "upload_bytes_per_second": 5346138.096639047,
+    "generation_bytes_per_second": 38961.11096808633,
+    "upload_bytes_per_second": 5314254.954739658,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 8.461122625927251,
+    "estimated_full_shards_per_hour": 8.3601474455065,
     "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": 1.4119799629988847,
-      "bytes": 11493188,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.9430077399993024,
       "bytes": 1135014,
@@ -562,6 +556,12 @@
       "bytes": 984851,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.161854639001831,
+      "bytes": 7666603,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -337,3 +337,4 @@
 {"step": 718700, "loss": 0.5621911287307739, "lr": 8e-05, "grad_norm": 0.22148633925331962, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105197791540, "tokens_per_second": 151198.74785465543, "data_wait_s": 9.552997653372586e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 718800, "loss": 0.5639420747756958, "lr": 8e-05, "grad_norm": 0.21778007256652135, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105212535088, "tokens_per_second": 154353.76146714773, "data_wait_s": 1.52729990077205e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 718900, "loss": 0.577467679977417, "lr": 8e-05, "grad_norm": 0.22120885357620004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105227313897, "tokens_per_second": 154028.4385952602, "data_wait_s": 1.5370002074632794e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 719000, "loss": 0.5348767638206482, "lr": 8e-05, "grad_norm": 0.21006133985819403, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105242001386, "tokens_per_second": 152235.51998967447, "data_wait_s": 9.997995221056044e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c690ded80fb21bc4e325e47a7af5219759c891716f4fd534dc132c748796b72b
-size 722865
+oid sha256:ba9a723cec361deeb5206852d15542dc6309302b813df07b9486b36c0875490d
+size 724983