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": 76035.14064129611,
-      "required_pages_per_second_upper_estimate": 0.0052602874553978884
+      "required_token_positions_per_second": 76832.94365758017,
+      "required_pages_per_second_upper_estimate": 0.005315481319222458
     },
     {
       "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": 9504.392580162013,
-      "required_pages_per_second_upper_estimate": 0.003979290699027498
+      "required_token_positions_per_second": 9604.11795719752,
+      "required_pages_per_second_upper_estimate": 0.0040210436318895825
     },
     {
       "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": 19008.785160324027,
-      "required_pages_per_second_upper_estimate": 0.003959028653520163
+      "required_token_positions_per_second": 19208.23591439504,
+      "required_pages_per_second_upper_estimate": 0.00400056898572306
     },
     {
       "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": 28513.177740486037,
-      "required_pages_per_second_upper_estimate": 0.0017093214684406081
+      "required_token_positions_per_second": 28812.35387159256,
+      "required_pages_per_second_upper_estimate": 0.0017272566206849426
     },
     {
       "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": 28513.177740486037,
-      "required_pages_per_second_upper_estimate": 0.0022167741531083033
+      "required_token_positions_per_second": 28812.35387159256,
+      "required_pages_per_second_upper_estimate": 0.002240033781365107
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28513.177740486037,
-      "required_pages_per_second_upper_estimate": 0.365736428989957
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28812.35387159256,
+      "required_pages_per_second_upper_estimate": 0.3695739391694894
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4004502575389779,
-    "producer_p95_seconds": 0.2728309980011545,
-    "gpu_update_seconds": 0.9696569162385024,
-    "producer_batches_per_second": 2.49718905450489,
-    "gpu_batches_per_second": 1.0312925976738294,
-    "producer_headroom": 2.4214166378557533,
+    "producer_batch_seconds": 0.2938680931798672,
+    "producer_p95_seconds": 0.28481914599979063,
+    "gpu_update_seconds": 0.959588380845879,
+    "producer_batches_per_second": 3.4028872926600178,
+    "gpu_batches_per_second": 1.0421134936195227,
+    "producer_headroom": 3.2653711073646434,
     "sustainable": true,
-    "buffer_seconds": 60.118728806787146,
+    "buffer_seconds": 60.45406799329038,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344415.9270180628,
-    "upload_bytes_per_second": 4819060.377032589,
+    "generation_bytes_per_second": 339885.2972762095,
+    "upload_bytes_per_second": 4832319.261631986,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.90364034563459,
+    "estimated_full_shards_per_hour": 72.93147267069544,
     "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": 1.2628126730014628,
-      "bytes": 9956766,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 1.1540077350000502,
       "bytes": 2614009,
@@ -562,6 +556,12 @@
       "bytes": 5662165,
       "pages": 21,
       "shards": 2
+    },
+    {
+      "seconds": 1.381297531999735,
+      "bytes": 11693702,
+      "pages": 44,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -287,3 +287,4 @@
 {"step": 788700, "loss": 0.530183732509613, "lr": 3e-05, "grad_norm": 0.2660127074293508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115443044566, "tokens_per_second": 151238.5717173458, "data_wait_s": 1.8304999684914947e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788800, "loss": 0.5539882779121399, "lr": 3e-05, "grad_norm": 0.26287225374246087, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115457688429, "tokens_per_second": 143153.21057854767, "data_wait_s": 1.064100069925189e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 788900, "loss": 0.5504850745201111, "lr": 3e-05, "grad_norm": 0.26454972272552435, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115472317052, "tokens_per_second": 151626.21056267782, "data_wait_s": 1.154200072051026e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 789000, "loss": 0.5050582885742188, "lr": 3e-05, "grad_norm": 0.2650915935230674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115486997166, "tokens_per_second": 151954.43834289283, "data_wait_s": 9.687999408924952e-06, "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 sha256:623690c0dcb6630023812f4756176d6943d31b44ac70e3bc460f4b97e05cf675
-size 616182
+oid sha256:f180ff8800a4e5f315de3ec831fa8c5bb3039569819ccdba962a1bcf5aad26a8
+size 618300