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": 76468.29400915361,
-      "required_pages_per_second_upper_estimate": 0.005290254010440554
+      "required_token_positions_per_second": 75571.35880936122,
+      "required_pages_per_second_upper_estimate": 0.005228201952142523
     },
     {
       "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": 9558.536751144202,
-      "required_pages_per_second_upper_estimate": 0.004001959732758878
+      "required_token_positions_per_second": 9446.419851170152,
+      "required_pages_per_second_upper_estimate": 0.0039550187280068495
     },
     {
       "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": 19117.073502288404,
-      "required_pages_per_second_upper_estimate": 0.003981582259395724
+      "required_token_positions_per_second": 18892.839702340305,
+      "required_pages_per_second_upper_estimate": 0.003934880272309501
     },
     {
       "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": 28675.610253432606,
-      "required_pages_per_second_upper_estimate": 0.0017190590495716686
+      "required_token_positions_per_second": 28339.25955351046,
+      "required_pages_per_second_upper_estimate": 0.0016988953386891175
     },
     {
       "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": 28675.610253432606,
-      "required_pages_per_second_upper_estimate": 0.002229402566524781
+      "required_token_positions_per_second": 28339.25955351046,
+      "required_pages_per_second_upper_estimate": 0.0022032527790562103
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28675.610253432606,
-      "required_pages_per_second_upper_estimate": 0.3678199388595914
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28339.25955351046,
+      "required_pages_per_second_upper_estimate": 0.3635055932262344
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10752830384404888,
-    "producer_p95_seconds": 0.2976577150038793,
-    "gpu_update_seconds": 0.9641643109126302,
-    "producer_batches_per_second": 9.29987700215495,
-    "gpu_batches_per_second": 1.0371676162265844,
-    "producer_headroom": 8.966609501354945,
+    "producer_batch_seconds": 0.1143994510780999,
+    "producer_p95_seconds": 0.3118377549981233,
+    "gpu_update_seconds": 0.9756077059033523,
+    "producer_batches_per_second": 8.741300684365218,
+    "gpu_batches_per_second": 1.0250021539898169,
+    "producer_headroom": 8.528080307284954,
     "sustainable": true,
-    "buffer_seconds": 60.742351587495705,
+    "buffer_seconds": 60.48767776600784,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362828.30826129735,
+    "generation_bytes_per_second": 363003.40533357725,
     "upload_bytes_per_second": 4912510.750269098,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.854508742134,
+    "estimated_full_shards_per_hour": 77.89208049779404,
     "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-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -349,3 +349,4 @@
 {"step": 794900, "loss": 0.5389185547828674, "lr": 3e-05, "grad_norm": 0.2768717896283943, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116350658479, "tokens_per_second": 151620.15939716657, "data_wait_s": 1.3008997484575957e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 795000, "loss": 0.4829784333705902, "lr": 3e-05, "grad_norm": 0.26054928311181613, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116365386659, "tokens_per_second": 152847.1465283554, "data_wait_s": 1.2873999367002398e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 795100, "loss": 0.5425436496734619, "lr": 3e-05, "grad_norm": 0.25600954343076604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116380042872, "tokens_per_second": 138664.39279502365, "data_wait_s": 7.835995347704738e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 795200, "loss": 0.5257145762443542, "lr": 3e-05, "grad_norm": 0.2820711886502756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 116394635879, "tokens_per_second": 151854.3283704057, "data_wait_s": 1.1088995961472392e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8130f31eed76c912f7a7a8a8f30dd3e33aeced5e3e4cf5b122f3ef434e11a9ef
-size 749064
+oid sha256:a69ec869142b5fd12b3ab3b229b60df83eb647a6a22d07d62295f7bd1a9c3fbb
+size 751182