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": 77223.14094821885,
-      "required_pages_per_second_upper_estimate": 0.005342476073171293
+      "required_token_positions_per_second": 76375.86555666526,
+      "required_pages_per_second_upper_estimate": 0.005283859595633852
     },
     {
       "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": 9652.892618527356,
-      "required_pages_per_second_upper_estimate": 0.0040414645640576355
+      "required_token_positions_per_second": 9546.983194583157,
+      "required_pages_per_second_upper_estimate": 0.003997122499892466
     },
     {
       "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": 19305.78523705471,
-      "required_pages_per_second_upper_estimate": 0.0040208859370844344
+      "required_token_positions_per_second": 19093.966389166315,
+      "required_pages_per_second_upper_estimate": 0.003976769657107944
     },
     {
       "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": 28958.677855582067,
-      "required_pages_per_second_upper_estimate": 0.0017360285201013265
+      "required_token_positions_per_second": 28640.949583749472,
+      "required_pages_per_second_upper_estimate": 0.001716981195347943
     },
     {
       "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": 28958.677855582067,
-      "required_pages_per_second_upper_estimate": 0.0022514098275091044
+      "required_token_positions_per_second": 28640.949583749472,
+      "required_pages_per_second_upper_estimate": 0.002226707851913092
     },
     {
       "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": 28958.677855582067,
-      "required_pages_per_second_upper_estimate": 0.37145082612565344
+      "required_token_positions_per_second": 28640.949583749472,
+      "required_pages_per_second_upper_estimate": 0.36737534900462376
     }
   ],
   "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": 0.08367420195295949,
-    "producer_p95_seconds": 0.2877474710003298,
-    "gpu_update_seconds": 0.9547397204348049,
-    "producer_batches_per_second": 11.9511148796159,
-    "gpu_batches_per_second": 1.0474058830867357,
-    "producer_headroom": 11.410204079048722,
+    "producer_batch_seconds": 0.3354286800235968,
+    "producer_p95_seconds": 0.2997576189991378,
+    "gpu_update_seconds": 0.9653311221108095,
+    "producer_batches_per_second": 2.981259682176407,
+    "gpu_batches_per_second": 1.0359139751066795,
+    "producer_headroom": 2.8779027542990665,
     "sustainable": true,
-    "buffer_seconds": 60.148602387392714,
+    "buffer_seconds": 60.815860692981,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 514926.8373027213,
-    "upload_bytes_per_second": 5224895.868396162,
+    "generation_bytes_per_second": 459987.4288282087,
+    "upload_bytes_per_second": 5088586.784185092,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.49131240187863,
+    "estimated_full_shards_per_hour": 98.70259426722237,
     "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": 1.2442956449995108,
-      "bytes": 13618701,
-      "pages": 45,
-      "shards": 4
-    },
     {
       "seconds": 0.9636815740013844,
       "bytes": 1058213,
@@ -562,6 +556,12 @@
       "bytes": 7481151,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.9575125699993805,
+      "bytes": 1008516,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -278,3 +278,4 @@
 {"step": 632200, "loss": 0.5619529485702515, "lr": 0.0001, "grad_norm": 0.16172479192987982, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92537393479, "tokens_per_second": 151958.09665879665, "data_wait_s": 1.5820001863176003e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 632300, "loss": 0.5189945101737976, "lr": 0.0001, "grad_norm": 0.16577218858353351, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92552137809, "tokens_per_second": 152821.69150754684, "data_wait_s": 1.4707002264913172e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 632400, "loss": 0.5342972278594971, "lr": 0.0001, "grad_norm": 0.1566162574160097, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92566689615, "tokens_per_second": 150764.29151361203, "data_wait_s": 1.428700124961324e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 632500, "loss": 0.5851893424987793, "lr": 0.0001, "grad_norm": 0.18170721393562317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92581282248, "tokens_per_second": 152111.3946410792, "data_wait_s": 1.4494002243736759e-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 sha256:5b162c5f7957ebbdf198a75f342a2269d0573a701c957fecaed5d878d0768978
-size 597903
+oid sha256:1f8ae5067f1bc5adc7e744f747800382d74373cd7126cbdf896badc37a984dfa
+size 600021