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": 77304.71381408648,
-      "required_pages_per_second_upper_estimate": 0.005348119473307142
+      "required_token_positions_per_second": 77445.08511839515,
+      "required_pages_per_second_upper_estimate": 0.005357830685845515
     },
     {
       "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": 9663.08922676081,
-      "required_pages_per_second_upper_estimate": 0.00404573367099559
+      "required_token_positions_per_second": 9680.635639799393,
+      "required_pages_per_second_upper_estimate": 0.004053079987724075
     },
     {
       "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": 19326.17845352162,
-      "required_pages_per_second_upper_estimate": 0.004025133306269207
+      "required_token_positions_per_second": 19361.271279598786,
+      "required_pages_per_second_upper_estimate": 0.004032442216481023
     },
     {
       "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": 28989.26768028243,
-      "required_pages_per_second_upper_estimate": 0.0017378623333841555
+      "required_token_positions_per_second": 29041.90691939818,
+      "required_pages_per_second_upper_estimate": 0.0017410179753936804
     },
     {
       "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": 28989.26768028243,
-      "required_pages_per_second_upper_estimate": 0.0022537880518291387
+      "required_token_positions_per_second": 29041.90691939818,
+      "required_pages_per_second_upper_estimate": 0.0022578805211348448
     },
     {
       "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": 28989.26768028243,
-      "required_pages_per_second_upper_estimate": 0.3718431995521149
+      "required_token_positions_per_second": 29041.90691939818,
+      "required_pages_per_second_upper_estimate": 0.3725183991918803
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,17 +155,17 @@
   "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": 0.04928668579687212,
-    "producer_p95_seconds": 0.1342790219998733,
-    "gpu_update_seconds": 0.9537322675731226,
-    "producer_batches_per_second": 20.289455130364292,
-    "gpu_batches_per_second": 1.048512285889845,
-    "producer_headroom": 19.350708049305464,
+    "producer_batch_seconds": 0.04851462940623996,
+    "producer_p95_seconds": 0.12414443499983463,
+    "gpu_update_seconds": 0.9520036021303017,
+    "producer_batches_per_second": 20.61233925186657,
+    "gpu_batches_per_second": 1.0504161935546217,
+    "producer_headroom": 19.62302121610878,
     "sustainable": true,
-    "buffer_seconds": 60.08513285710672,
+    "buffer_seconds": 60.928230536339306,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 69,

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -15,3 +15,4 @@
 {"step": 761500, "loss": 0.5549948215484619, "lr": 3e-05, "grad_norm": 0.25229013580404586, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111461313256, "tokens_per_second": 152066.6605562098, "data_wait_s": 8.672999911141233e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 761600, "loss": 0.5470552444458008, "lr": 3e-05, "grad_norm": 0.25005745380968136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111475755654, "tokens_per_second": 151434.22608220068, "data_wait_s": 1.94229999124218e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 761700, "loss": 0.5007930397987366, "lr": 3e-05, "grad_norm": 0.23310555419277346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111490271322, "tokens_per_second": 152385.58246989583, "data_wait_s": 1.2020999747619499e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 761800, "loss": 0.5855912566184998, "lr": 3e-05, "grad_norm": 0.27354794981320624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 111505032369, "tokens_per_second": 154600.75998703312, "data_wait_s": 1.6495000181748765e-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 sha256:440bc98af3f7bdd1673654bc967c0b4b2bd94f1006d221ae7ea1e4316e560e14
-size 36171
+oid sha256:e3b9d56a83bf16ea78baaed36a604672357da1d60f7f243ed33d1bf265b100b4
+size 38289