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": 76785.34468962895,
-      "required_pages_per_second_upper_estimate": 0.005312188312174771
+      "required_token_positions_per_second": 76488.39446077497,
+      "required_pages_per_second_upper_estimate": 0.005291644606323193
     },
     {
       "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": 9598.16808620362,
-      "required_pages_per_second_upper_estimate": 0.0040185525451518714
+      "required_token_positions_per_second": 9561.049307596872,
+      "required_pages_per_second_upper_estimate": 0.004003011687677466
     },
     {
       "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": 19196.33617240724,
-      "required_pages_per_second_upper_estimate": 0.0039980905832842995
+      "required_token_positions_per_second": 19122.098615193743,
+      "required_pages_per_second_upper_estimate": 0.0039826288578927675
     },
     {
       "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": 28794.504258610854,
-      "required_pages_per_second_upper_estimate": 0.00172618656364147
+      "required_token_positions_per_second": 28683.147922790613,
+      "required_pages_per_second_upper_estimate": 0.0017195109213403263
     },
     {
       "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": 28794.504258610854,
-      "required_pages_per_second_upper_estimate": 0.0022386460524679293
+      "required_token_positions_per_second": 28683.147922790613,
+      "required_pages_per_second_upper_estimate": 0.0022299885871626627
     },
     {
       "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": 28794.504258610854,
-      "required_pages_per_second_upper_estimate": 0.3693449834995812
+      "required_token_positions_per_second": 28683.147922790613,
+      "required_pages_per_second_upper_estimate": 0.36791662398879715
     }
   ],
   "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": 0.10534007390637612,
-    "producer_p95_seconds": 0.22015652299887734,
-    "gpu_update_seconds": 0.9601832263436868,
-    "producer_batches_per_second": 9.493063398538883,
-    "gpu_batches_per_second": 1.0414678912981357,
-    "producer_headroom": 9.11508024189423,
+    "producer_batch_seconds": 0.24894184325793844,
+    "producer_p95_seconds": 0.21085809000214795,
+    "gpu_update_seconds": 0.9639109373358521,
+    "producer_batches_per_second": 4.017002472998726,
+    "gpu_batches_per_second": 1.0374402460500078,
+    "producer_headroom": 3.872032619028638,
     "sustainable": true,
-    "buffer_seconds": 60.49154325965227,
+    "buffer_seconds": 60.72638905215868,
     "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": 556179.130581579,
-    "upload_bytes_per_second": 5168575.3491557585,
+    "generation_bytes_per_second": 512662.89842170116,
+    "upload_bytes_per_second": 5242928.3085958855,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 119.34309423528222,
+    "estimated_full_shards_per_hour": 110.0055238198116,
     "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": 1.0847412589992018,
-      "bytes": 1126127,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.2304532270009076,
       "bytes": 11321849,
@@ -562,6 +556,12 @@
       "bytes": 11737942,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1044216970003617,
+      "bytes": 7366106,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -163,3 +163,4 @@
 {"step": 620700, "loss": 0.5945292115211487, "lr": 0.0001, "grad_norm": 0.17012109593468946, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90853791441, "tokens_per_second": 152786.06384667708, "data_wait_s": 1.4075998478801921e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 620800, "loss": 0.5377899408340454, "lr": 0.0001, "grad_norm": 0.17761723371324462, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90868425634, "tokens_per_second": 152212.34361118838, "data_wait_s": 1.4498000382445753e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 620900, "loss": 0.5788757801055908, "lr": 0.0001, "grad_norm": 0.1803991948934275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90883023139, "tokens_per_second": 151778.5397582285, "data_wait_s": 9.416999091627076e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 621000, "loss": 0.5698689818382263, "lr": 0.0001, "grad_norm": 0.15833012025703624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90897686835, "tokens_per_second": 152322.75978387456, "data_wait_s": 1.432100179954432e-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 sha256:6af4b4293323f2bc0e3d2f1fe7bf269da46ea68c1763a0e825a86fbd7d3166a3
-size 352767
+oid sha256:a49de46ecb5b2386e37402b3b888cc7de070a9225b0e44ad1a4e345b4b2665a4
+size 354885