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": 76889.27697536902,
-      "required_pages_per_second_upper_estimate": 0.0053193785888583965
+      "required_token_positions_per_second": 77060.14448798755,
+      "required_pages_per_second_upper_estimate": 0.005331199599848598
     },
     {
       "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": 9611.159621921128,
-      "required_pages_per_second_upper_estimate": 0.004023991829862676
+      "required_token_positions_per_second": 9632.518060998444,
+      "required_pages_per_second_upper_estimate": 0.0040329341622894225
     },
     {
       "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": 19222.319243842256,
-      "required_pages_per_second_upper_estimate": 0.004003502171844541
+      "required_token_positions_per_second": 19265.036121996887,
+      "required_pages_per_second_upper_estimate": 0.004012398971044319
     },
     {
       "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": 28833.478865763384,
-      "required_pages_per_second_upper_estimate": 0.0017285230318294848
+      "required_token_positions_per_second": 28897.55418299533,
+      "required_pages_per_second_upper_estimate": 0.0017323642492601957
     },
     {
       "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": 28833.478865763384,
-      "required_pages_per_second_upper_estimate": 0.0022416761567428593
+      "required_token_positions_per_second": 28897.55418299533,
+      "required_pages_per_second_upper_estimate": 0.002246657730820108
     },
     {
       "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": 28833.478865763384,
-      "required_pages_per_second_upper_estimate": 0.36984490791246116
+      "required_token_positions_per_second": 28897.55418299533,
+      "required_pages_per_second_upper_estimate": 0.3706667972831971
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.15167661192970172,
-    "producer_p95_seconds": 0.35507209300703835,
-    "gpu_update_seconds": 0.9588853335637203,
-    "producer_batches_per_second": 6.5929742712309185,
-    "gpu_batches_per_second": 1.0428775631424834,
-    "producer_headroom": 6.321906333246285,
+    "producer_batch_seconds": 0.13888229553117526,
+    "producer_p95_seconds": 0.363497867998376,
+    "gpu_update_seconds": 0.9567591715519431,
+    "producer_batches_per_second": 7.200341815890619,
+    "gpu_batches_per_second": 1.0451951021048658,
+    "producer_headroom": 6.888993070662322,
     "sustainable": true,
-    "buffer_seconds": 60.40977601451438,
+    "buffer_seconds": 60.275827807772416,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361522.7692592011,
+    "generation_bytes_per_second": 360738.03359167324,
     "upload_bytes_per_second": 4634033.565828152,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.57437046367669,
+    "estimated_full_shards_per_hour": 77.40598445713661,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -403,3 +403,4 @@
 {"step": 644700, "loss": 0.5330870151519775, "lr": 0.0001, "grad_norm": 0.1697825474437093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94367436580, "tokens_per_second": 149924.6990900405, "data_wait_s": 1.9946004613302648e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 644800, "loss": 0.5745224952697754, "lr": 0.0001, "grad_norm": 0.17889248196152127, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94381940687, "tokens_per_second": 144931.13760888163, "data_wait_s": 7.866998203098774e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 644900, "loss": 0.48915520310401917, "lr": 0.0001, "grad_norm": 0.18857827802658592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94396571499, "tokens_per_second": 151698.3424844779, "data_wait_s": 1.4887998986523598e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 645000, "loss": 0.5380465984344482, "lr": 0.0001, "grad_norm": 0.1812405295203612, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94411289736, "tokens_per_second": 151737.97812397123, "data_wait_s": 1.7938000382855535e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd020c742fedb96a971087ff35fd517a439d7047a4018993c7eb98bc471883ed
-size 864219
+oid sha256:828cc8763870c1af05afdd3663129dae74f8909f66bc5fe9e2fdbd43bfc38f5d
+size 866337