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": 75992.32621400141,
-      "required_pages_per_second_upper_estimate": 0.005257325453974484
+      "required_token_positions_per_second": 75722.9040660672,
+      "required_pages_per_second_upper_estimate": 0.005238686204634884
     },
     {
       "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": 9499.040776750177,
-      "required_pages_per_second_upper_estimate": 0.003977050010697325
+      "required_token_positions_per_second": 9465.3630082584,
+      "required_pages_per_second_upper_estimate": 0.003962949832301597
     },
     {
       "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": 18998.081553500353,
-      "required_pages_per_second_upper_estimate": 0.003956799374491886
+      "required_token_positions_per_second": 18930.7260165168,
+      "required_pages_per_second_upper_estimate": 0.003942770992423172
     },
     {
       "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": 28497.122330250528,
-      "required_pages_per_second_upper_estimate": 0.0017083589711121812
+      "required_token_positions_per_second": 28396.0890247752,
+      "required_pages_per_second_upper_estimate": 0.0017023021787178603
     },
     {
       "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": 28497.122330250528,
-      "required_pages_per_second_upper_estimate": 0.0022155259155827787
+      "required_token_positions_per_second": 28396.0890247752,
+      "required_pages_per_second_upper_estimate": 0.0022076710204806172
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28497.122330250528,
-      "required_pages_per_second_upper_estimate": 0.36553048742641225
+      "required_token_positions_per_second": 28396.0890247752,
+      "required_pages_per_second_upper_estimate": 0.3642345406648863
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.15895775521886435,
-    "producer_p95_seconds": 0.38224408599489834,
-    "gpu_update_seconds": 0.9702032254200923,
-    "producer_batches_per_second": 6.290979629292883,
-    "gpu_batches_per_second": 1.0307118898383438,
-    "producer_headroom": 6.103528727392051,
+    "producer_batch_seconds": 0.1257561896092625,
+    "producer_p95_seconds": 0.3574241579990485,
+    "gpu_update_seconds": 0.9736552092042499,
+    "producer_batches_per_second": 7.951894877755946,
+    "gpu_batches_per_second": 1.0270576180835937,
+    "producer_headroom": 7.742403870771669,
     "sustainable": true,
-    "buffer_seconds": 60.15259997604572,
+    "buffer_seconds": 60.36662297066349,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 496233.626620104,
-    "upload_bytes_per_second": 4575503.194682716,
+    "generation_bytes_per_second": 523263.4243114643,
+    "upload_bytes_per_second": 4519218.487224701,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.48018454506244,
+    "estimated_full_shards_per_hour": 112.28014990814158,
     "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": 1.49954066099599,
-      "bytes": 11232631,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.3497194699957618,
       "bytes": 2429729,
@@ -562,6 +556,12 @@
       "bytes": 2343346,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.5891393990023062,
+      "bytes": 6271305,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -432,3 +432,4 @@
 {"step": 803200, "loss": 0.5094794631004333, "lr": 3e-05, "grad_norm": 0.26051532754282786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117565077044, "tokens_per_second": 150728.77412666593, "data_wait_s": 9.891002264339477e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 803300, "loss": 0.5157017707824707, "lr": 3e-05, "grad_norm": 0.2845687018671663, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117579791596, "tokens_per_second": 152538.1604435676, "data_wait_s": 1.0247997124679387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 803400, "loss": 0.5423639416694641, "lr": 3e-05, "grad_norm": 0.2784970674390396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117594576659, "tokens_per_second": 153697.22625140508, "data_wait_s": 8.11300560599193e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 803500, "loss": 0.5135219693183899, "lr": 3e-05, "grad_norm": 0.28931642332439106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117609406982, "tokens_per_second": 154380.1701985896, "data_wait_s": 2.8823000320699066e-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:bacf0f14004697617f85dd4c84cc642841f491b5e577ea914c0307835cea1f5a
-size 925641
+oid sha256:7f198379956ffd04885fb23c56d7cfbf659449832ad6aebcb3ba5d1526e166cc
+size 927759