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": 77366.06903308831,
-      "required_pages_per_second_upper_estimate": 0.005352364169721411
+      "required_token_positions_per_second": 76854.60630171224,
+      "required_pages_per_second_upper_estimate": 0.005316979991207773
     },
     {
       "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": 9670.758629136039,
-      "required_pages_per_second_upper_estimate": 0.004048944689614764
+      "required_token_positions_per_second": 9606.82578771403,
+      "required_pages_per_second_upper_estimate": 0.004022177343980914
     },
     {
       "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": 19341.517258272077,
-      "required_pages_per_second_upper_estimate": 0.004028327974787244
+      "required_token_positions_per_second": 19213.65157542806,
+      "required_pages_per_second_upper_estimate": 0.004001696925095652
     },
     {
       "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": 29012.275887408116,
-      "required_pages_per_second_upper_estimate": 0.0017392416402698422
+      "required_token_positions_per_second": 28820.47736314209,
+      "required_pages_per_second_upper_estimate": 0.0017277436115994834
     },
     {
       "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": 29012.275887408116,
-      "required_pages_per_second_upper_estimate": 0.002255576838730753
+      "required_token_positions_per_second": 28820.47736314209,
+      "required_pages_per_second_upper_estimate": 0.0022406653471016198
     },
     {
       "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": 29012.275887408116,
-      "required_pages_per_second_upper_estimate": 0.37213832412883513
+      "required_token_positions_per_second": 28820.47736314209,
+      "required_pages_per_second_upper_estimate": 0.3696781385967611
     }
   ],
   "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": 0.09671359894530696,
-    "producer_p95_seconds": 0.1813451210000494,
-    "gpu_update_seconds": 0.9529759094838803,
-    "producer_batches_per_second": 10.339807544185337,
-    "gpu_batches_per_second": 1.0493444693072957,
-    "producer_headroom": 9.85358749830831,
+    "producer_batch_seconds": 0.28573915822656204,
+    "producer_p95_seconds": 0.1837381989998903,
+    "gpu_update_seconds": 0.9593179062106186,
+    "producer_batches_per_second": 3.499695338246576,
+    "gpu_batches_per_second": 1.0424073120349424,
+    "producer_headroom": 3.357320404261768,
     "sustainable": true,
-    "buffer_seconds": 60.03748229748446,
+    "buffer_seconds": 60.43702809126897,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 387728.65591454133,
-    "upload_bytes_per_second": 5084444.9503038945,
+    "generation_bytes_per_second": 384630.08557053364,
+    "upload_bytes_per_second": 5114015.67570297,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.19754369809323,
+    "estimated_full_shards_per_hour": 82.53266263329513,
     "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": [
@@ -316,6 +316,12 @@
       "bytes": 7151255,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1764370720002262,
+      "bytes": 6881233,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -62,3 +62,4 @@
 {"step": 766200, "loss": 0.564631998538971, "lr": 3e-05, "grad_norm": 0.280078313192073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112149023207, "tokens_per_second": 141144.55896482535, "data_wait_s": 9.847000001173e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 766300, "loss": 0.5835681557655334, "lr": 3e-05, "grad_norm": 0.27350498943782986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112163701144, "tokens_per_second": 153281.3142035325, "data_wait_s": 9.481000233790837e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 766400, "loss": 0.5553596019744873, "lr": 3e-05, "grad_norm": 0.24810549232862186, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112178338515, "tokens_per_second": 152014.75280486417, "data_wait_s": 8.891999641491566e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 766500, "loss": 0.5904669165611267, "lr": 3e-05, "grad_norm": 0.2738577587243592, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112192956436, "tokens_per_second": 152648.770074057, "data_wait_s": 1.1748999895644374e-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:c594f4b6c537c6392905fab4354a51a87a954af6f38f1a1e0ef02263b0ca7cd5
-size 136500
+oid sha256:a621db6e9fdf9e4221db23e685b3ae35db4e0c0cc2d64a4e7f4b4e4b37acf0f6
+size 138618