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": 76682.5465948514,
-      "required_pages_per_second_upper_estimate": 0.005305076501453618
+      "required_token_positions_per_second": 75828.56187006152,
+      "required_pages_per_second_upper_estimate": 0.005245995856675092
     },
     {
       "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": 9585.318324356425,
-      "required_pages_per_second_upper_estimate": 0.004013172618199992
+      "required_token_positions_per_second": 9478.57023375769,
+      "required_pages_per_second_upper_estimate": 0.003968479421819918
     },
     {
       "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": 19170.63664871285,
-      "required_pages_per_second_upper_estimate": 0.003992738050240775
+      "required_token_positions_per_second": 18957.14046751538,
+      "required_pages_per_second_upper_estimate": 0.003948272425970259
     },
     {
       "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": 28755.954973069278,
-      "required_pages_per_second_upper_estimate": 0.0017238755928346034
+      "required_token_positions_per_second": 28435.710701273067,
+      "required_pages_per_second_upper_estimate": 0.0017046774377251104
     },
     {
       "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": 28755.954973069278,
-      "required_pages_per_second_upper_estimate": 0.0022356490150774594
+      "required_token_positions_per_second": 28435.710701273067,
+      "required_pages_per_second_upper_estimate": 0.0022107514315509893
     },
     {
       "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": 28755.954973069278,
-      "required_pages_per_second_upper_estimate": 0.3688505146556519
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28435.710701273067,
+      "required_pages_per_second_upper_estimate": 0.36474276498855923
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12267156260907086,
-    "producer_p95_seconds": 0.3505541720005567,
-    "gpu_update_seconds": 0.9614704163326551,
-    "producer_batches_per_second": 8.15184855178535,
-    "gpu_batches_per_second": 1.040073602903258,
-    "producer_headroom": 7.8377612209658105,
+    "producer_batch_seconds": 0.4929478838204773,
+    "producer_p95_seconds": 0.3399224940003478,
+    "gpu_update_seconds": 0.9722985400453592,
+    "producer_batches_per_second": 2.028612015229143,
+    "gpu_batches_per_second": 1.0284906937671103,
+    "producer_headroom": 1.9724165007257701,
     "sustainable": true,
-    "buffer_seconds": 60.57263622895727,
+    "buffer_seconds": 60.28250948281227,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339265.5788432134,
-    "upload_bytes_per_second": 4556795.149347473,
+    "generation_bytes_per_second": 298742.058493403,
+    "upload_bytes_per_second": 4523390.495362184,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.79849552128125,
+    "estimated_full_shards_per_hour": 64.10309139348571,
     "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": 1.2226353290025145,
-      "bytes": 7620412,
-      "pages": 40,
-      "shards": 3
-    },
     {
       "seconds": 1.2109157329978188,
       "bytes": 5386870,
@@ -562,6 +556,12 @@
       "bytes": 2378178,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.5392421450014808,
+      "bytes": 5890105,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -410,3 +410,4 @@
 {"step": 801000, "loss": 0.5692843794822693, "lr": 3e-05, "grad_norm": 0.2703526957225969, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117242932024, "tokens_per_second": 151489.24277070843, "data_wait_s": 9.110997780226171e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 801100, "loss": 0.5213542580604553, "lr": 3e-05, "grad_norm": 0.2853517031856212, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117257534817, "tokens_per_second": 151063.8605249472, "data_wait_s": 1.109600270865485e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 801200, "loss": 0.5780267119407654, "lr": 3e-05, "grad_norm": 0.2922971514564205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117272170109, "tokens_per_second": 130965.90942947168, "data_wait_s": 9.925999620463699e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 801300, "loss": 0.5488598346710205, "lr": 3e-05, "grad_norm": 0.2763455230250154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117286883998, "tokens_per_second": 152350.59076349984, "data_wait_s": 1.4092001947574317e-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 sha256:4961bc36414da18ccd64a7bd64905ce5b13f6180c948bb70072d4a6a60a5ebf9
-size 879045
+oid sha256:988ea3bb327eb94d569e7afa5cfa1f4e22b3268e89f9e610c4107347740f9baf
+size 881163