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": 77141.50413822113,
-      "required_pages_per_second_upper_estimate": 0.005336828249232155
+      "required_token_positions_per_second": 76865.32356009017,
+      "required_pages_per_second_upper_estimate": 0.005317721436009825
     },
     {
       "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": 9642.688017277642,
-      "required_pages_per_second_upper_estimate": 0.004037192110610686
+      "required_token_positions_per_second": 9608.165445011271,
+      "required_pages_per_second_upper_estimate": 0.004022738230516043
     },
     {
       "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": 19285.376034555284,
-      "required_pages_per_second_upper_estimate": 0.004016635238430668
+      "required_token_positions_per_second": 19216.330890022542,
+      "required_pages_per_second_upper_estimate": 0.004002254955667405
     },
     {
       "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": 28928.064051832927,
-      "required_pages_per_second_upper_estimate": 0.0017341932693111374
+      "required_token_positions_per_second": 28824.49633503381,
+      "required_pages_per_second_upper_estimate": 0.0017279845428277774
     },
     {
       "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": 28928.064051832927,
-      "required_pages_per_second_upper_estimate": 0.0022490297389235006
+      "required_token_positions_per_second": 28824.49633503381,
+      "required_pages_per_second_upper_estimate": 0.002240977804488612
     },
     {
       "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": 28928.064051832927,
-      "required_pages_per_second_upper_estimate": 0.37105814512170093
+      "required_token_positions_per_second": 28824.49633503381,
+      "required_pages_per_second_upper_estimate": 0.3697296896529523
     }
   ],
   "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": 0.17490503389057466,
-    "producer_p95_seconds": 0.18144295299953228,
-    "gpu_update_seconds": 0.955750096185513,
-    "producer_batches_per_second": 5.717388332148445,
-    "gpu_batches_per_second": 1.0462986129858554,
-    "producer_headroom": 5.464394448380807,
+    "producer_batch_seconds": 0.06787018394526712,
+    "producer_p95_seconds": 0.1660532650003006,
+    "gpu_update_seconds": 0.9591841494345947,
+    "producer_batches_per_second": 14.73401045747032,
+    "gpu_batches_per_second": 1.0425526741548687,
+    "producer_headroom": 14.132629288409094,
     "sustainable": true,
-    "buffer_seconds": 60.21225605968732,
+    "buffer_seconds": 60.42860141437947,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435006.7282725615,
-    "upload_bytes_per_second": 5109631.980423423,
+    "generation_bytes_per_second": 462909.86043904105,
+    "upload_bytes_per_second": 5132989.268753572,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.34231744892725,
+    "estimated_full_shards_per_hour": 99.32968006018089,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 12463391,
       "pages": 43,
       "shards": 4
+    },
+    {
+      "seconds": 1.107387490000292,
+      "bytes": 5957600,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -52,3 +52,4 @@
 {"step": 690200, "loss": 0.5197702050209045, "lr": 8e-05, "grad_norm": 0.19167683174711828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101026325514, "tokens_per_second": 151782.48313986367, "data_wait_s": 1.764999979059212e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690300, "loss": 0.5576441884040833, "lr": 8e-05, "grad_norm": 0.20442363267717895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101040989737, "tokens_per_second": 153074.2715404991, "data_wait_s": 1.5315999917220324e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 690400, "loss": 0.5102057456970215, "lr": 8e-05, "grad_norm": 0.20372598556241328, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101055634316, "tokens_per_second": 152507.45737045284, "data_wait_s": 1.594600053067552e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 690500, "loss": 0.5640904903411865, "lr": 8e-05, "grad_norm": 0.22608068633271008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101070357154, "tokens_per_second": 154020.0132578584, "data_wait_s": 1.705900012893835e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d32c79d376967a15efdce7a39f520edc1d0eb074435fde2f2d045aaef2dc518
-size 115320
+oid sha256:27ad58c46db99f76aaffb59f10b16e5ee6219e5665b41b9a717e306b6bda1540
+size 117438