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": 76872.69737057458,
-      "required_pages_per_second_upper_estimate": 0.005318231573328733
+      "required_token_positions_per_second": 76734.94666234721,
+      "required_pages_per_second_upper_estimate": 0.005308701659707885
     },
     {
       "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": 9609.087171321822,
-      "required_pages_per_second_upper_estimate": 0.004023124138074437
+      "required_token_positions_per_second": 9591.868332793401,
+      "required_pages_per_second_upper_estimate": 0.0040159149699528236
     },
     {
       "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": 19218.174342643644,
-      "required_pages_per_second_upper_estimate": 0.004002638898233266
+      "required_token_positions_per_second": 19183.736665586803,
+      "required_pages_per_second_upper_estimate": 0.003995466438284942
     },
     {
       "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": 28827.26151396547,
-      "required_pages_per_second_upper_estimate": 0.0017281503110825466
+      "required_token_positions_per_second": 28775.604998380204,
+      "required_pages_per_second_upper_estimate": 0.0017250535818481422
     },
     {
       "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": 28827.26151396547,
-      "required_pages_per_second_upper_estimate": 0.0022411927849878117
+      "required_token_positions_per_second": 28775.604998380204,
+      "required_pages_per_second_upper_estimate": 0.002237176718114057
     },
     {
       "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": 28827.26151396547,
-      "required_pages_per_second_upper_estimate": 0.36976515839926977
+      "required_token_positions_per_second": 28775.604998380204,
+      "required_pages_per_second_upper_estimate": 0.3691025640817871
     }
   ],
   "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": 0.08948182104705893,
-    "producer_p95_seconds": 0.23560486099813716,
-    "gpu_update_seconds": 0.9590921422281415,
-    "producer_batches_per_second": 11.175454279971516,
-    "gpu_batches_per_second": 1.0426526878604407,
-    "producer_headroom": 10.718290385750535,
+    "producer_batch_seconds": 0.3246874721873496,
+    "producer_p95_seconds": 0.2461885650009208,
+    "gpu_update_seconds": 0.9608138560963817,
+    "producer_batches_per_second": 3.0798847681532497,
+    "gpu_batches_per_second": 1.0407843243048396,
+    "producer_headroom": 2.9591959604218343,
     "sustainable": true,
-    "buffer_seconds": 60.422804960372915,
+    "buffer_seconds": 60.53127293407205,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319125.1032919102,
-    "upload_bytes_per_second": 5028235.559843255,
+    "generation_bytes_per_second": 307011.26147657295,
+    "upload_bytes_per_second": 5019516.799788638,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.47681831424694,
+    "estimated_full_shards_per_hour": 65.87746985648052,
     "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": 1.2456822759995703,
-      "bytes": 12102575,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 0.872640274999867,
       "bytes": 1185922,
@@ -562,6 +556,12 @@
       "bytes": 2162409,
       "pages": 24,
       "shards": 2
+    },
+    {
+      "seconds": 1.2899976480002806,
+      "bytes": 11585751,
+      "pages": 45,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -189,3 +189,4 @@
 {"step": 778900, "loss": 0.5157540440559387, "lr": 3e-05, "grad_norm": 0.24314472441481658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114008826461, "tokens_per_second": 153012.47914859478, "data_wait_s": 1.2923999747727066e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 779000, "loss": 0.5144950151443481, "lr": 3e-05, "grad_norm": 0.2509918200475692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114023471031, "tokens_per_second": 149083.04023248106, "data_wait_s": 9.705003321869299e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 779100, "loss": 0.4944155514240265, "lr": 3e-05, "grad_norm": 0.25595977421221033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114038246442, "tokens_per_second": 153262.59007515403, "data_wait_s": 1.6082001820905134e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 779200, "loss": 0.5231711268424988, "lr": 3e-05, "grad_norm": 0.2526803621215525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114052935439, "tokens_per_second": 151390.63273724014, "data_wait_s": 1.5181001799646765e-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 sha256:7305996af6f703ff1aae54f324fb6fad082bb10e29aa1b898e425f517e2a8132
-size 407052
+oid sha256:62a42b7040c2fb4f3defe7bb969e66e053c66500d4d2cf6f52a2ee769da347a4
+size 409170