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": 77291.06542383185,
-      "required_pages_per_second_upper_estimate": 0.005347175246000698
+      "required_token_positions_per_second": 77337.79198135664,
+      "required_pages_per_second_upper_estimate": 0.005350407897670814
     },
     {
       "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": 9661.383177978982,
-      "required_pages_per_second_upper_estimate": 0.004045019383996988
+      "required_token_positions_per_second": 9667.22399766958,
+      "required_pages_per_second_upper_estimate": 0.004047464813231256
     },
     {
       "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": 19322.766355957963,
-      "required_pages_per_second_upper_estimate": 0.004024422656329761
+      "required_token_positions_per_second": 19334.44799533916,
+      "required_pages_per_second_upper_estimate": 0.004026855633747318
     },
     {
       "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": 28984.149533936947,
-      "required_pages_per_second_upper_estimate": 0.0017375555083257008
+      "required_token_positions_per_second": 29001.67199300874,
+      "required_pages_per_second_upper_estimate": 0.0017386059529918088
     },
     {
       "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": 28984.149533936947,
-      "required_pages_per_second_upper_estimate": 0.002253390138463124
+      "required_token_positions_per_second": 29001.67199300874,
+      "required_pages_per_second_upper_estimate": 0.0022547524325828034
     },
     {
       "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": 28984.149533936947,
-      "required_pages_per_second_upper_estimate": 0.3717775494662324
+      "required_token_positions_per_second": 29001.67199300874,
+      "required_pages_per_second_upper_estimate": 0.37200230875705464
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.4843429877890202,
-    "producer_p95_seconds": 0.21941636399969866,
-    "gpu_update_seconds": 0.9539006817373586,
-    "producer_batches_per_second": 2.0646525813554257,
-    "gpu_batches_per_second": 1.0483271677494557,
-    "producer_headroom": 1.969473504905738,
+    "producer_batch_seconds": 0.2906923827109864,
+    "producer_p95_seconds": 0.20526774199970532,
+    "gpu_update_seconds": 0.9533243464950897,
+    "producer_batches_per_second": 3.440062621091193,
+    "gpu_batches_per_second": 1.0489609372471331,
+    "producer_headroom": 3.2794954501539464,
     "sustainable": true,
-    "buffer_seconds": 60.095742949453594,
+    "buffer_seconds": 60.05943382919065,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342020.4375691451,
-    "upload_bytes_per_second": 3594227.05959672,
+    "generation_bytes_per_second": 328751.58465276496,
+    "upload_bytes_per_second": 3596231.5696296524,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.38962407403721,
+    "estimated_full_shards_per_hour": 70.54243712126933,
     "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": 1.3296988840002086,
-      "bytes": 7650823,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.2172214809997968,
       "bytes": 6191100,
@@ -562,6 +556,12 @@
       "bytes": 6164782,
       "pages": 33,
       "shards": 2
+    },
+    {
+      "seconds": 1.4110855310009356,
+      "bytes": 8184458,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -147,3 +147,4 @@
 {"step": 939700, "loss": 0.5374237895011902, "lr": 3e-05, "grad_norm": 0.3497322877010348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137545618407, "tokens_per_second": 154556.27126909661, "data_wait_s": 1.5528999938396737e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 939800, "loss": 0.5418662428855896, "lr": 3e-05, "grad_norm": 0.36800081730324674, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137560295837, "tokens_per_second": 147792.3388991289, "data_wait_s": 1.1980999261140823e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 939900, "loss": 0.49588456749916077, "lr": 3e-05, "grad_norm": 0.3256690265589337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137574954840, "tokens_per_second": 153068.80787155707, "data_wait_s": 1.1854999684146605e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 940000, "loss": 0.5351184010505676, "lr": 3e-05, "grad_norm": 0.3404110635583141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137589403422, "tokens_per_second": 150593.29881988783, "data_wait_s": 1.485900065745227e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3d86b45d96f00b7b4c942657ef4140c099fa68e051420993cd005491602d6d4
-size 317313
+oid sha256:c5c5e5d1b58661e1734800f4206b619cee440af04fb1f97ba5e68c9ad9b4c18e
+size 319431