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": 76092.56468731658,
-      "required_pages_per_second_upper_estimate": 0.0052642601841437296
+      "required_token_positions_per_second": 76839.57816559378,
+      "required_pages_per_second_upper_estimate": 0.005315940309881018
     },
     {
       "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": 9511.570585914573,
-      "required_pages_per_second_upper_estimate": 0.00398229598014229
+      "required_token_positions_per_second": 9604.947270699222,
+      "required_pages_per_second_upper_estimate": 0.00402139084813471
     },
     {
       "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": 19023.141171829146,
-      "required_pages_per_second_upper_estimate": 0.003962018632123197
+      "required_token_positions_per_second": 19209.894541398444,
+      "required_pages_per_second_upper_estimate": 0.004000914433986933
     },
     {
       "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": 28534.711757743717,
-      "required_pages_per_second_upper_estimate": 0.0017106123998946656
+      "required_token_positions_per_second": 28814.841812097668,
+      "required_pages_per_second_upper_estimate": 0.0017274057689193585
     },
     {
       "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": 28534.711757743717,
-      "required_pages_per_second_upper_estimate": 0.002218448328231956
+      "required_token_positions_per_second": 28814.841812097668,
+      "required_pages_per_second_upper_estimate": 0.0022402272077960855
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28534.711757743717,
-      "required_pages_per_second_upper_estimate": 0.3660126442419122
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28814.841812097668,
+      "required_pages_per_second_upper_estimate": 0.3696058517989465
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0757747054376523,
-    "producer_p95_seconds": 0.2678559670021059,
-    "gpu_update_seconds": 0.9689251545531004,
-    "producer_batches_per_second": 13.197015999261172,
-    "gpu_batches_per_second": 1.0320714611452444,
-    "producer_headroom": 12.78692076672387,
+    "producer_batch_seconds": 0.42169215313296604,
+    "producer_p95_seconds": 0.2812345700003789,
+    "gpu_update_seconds": 0.9595055277517512,
+    "producer_batches_per_second": 2.3713981694240456,
+    "gpu_batches_per_second": 1.0422034798935789,
+    "producer_headroom": 2.2753696520627558,
     "sustainable": true,
-    "buffer_seconds": 60.07335958229222,
+    "buffer_seconds": 60.44884824836033,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323558.1067559379,
-    "upload_bytes_per_second": 3206383.16920422,
+    "generation_bytes_per_second": 301234.8800606655,
+    "upload_bytes_per_second": 3238230.224910751,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.42803766258815,
+    "estimated_full_shards_per_hour": 64.63799287190412,
     "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": 1.6552812369991443,
-      "bytes": 6589515,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.7584444239982986,
       "bytes": 6116166,
@@ -562,6 +556,12 @@
       "bytes": 836473,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.9911568649986293,
+      "bytes": 11961950,
+      "pages": 42,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 829200, "loss": 0.5322349667549133, "lr": 3e-05, "grad_norm": 0.28532622734203844, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121371716667, "tokens_per_second": 140601.0131580074, "data_wait_s": 7.474998710677028e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 829300, "loss": 0.4824495017528534, "lr": 3e-05, "grad_norm": 0.27106723897553164, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121386298888, "tokens_per_second": 150986.97134923682, "data_wait_s": 1.4417997590498999e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 829400, "loss": 0.5459651947021484, "lr": 3e-05, "grad_norm": 0.28340184518973255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121401004413, "tokens_per_second": 153089.67534885972, "data_wait_s": 8.663999324198812e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 829500, "loss": 0.5509658455848694, "lr": 3e-05, "grad_norm": 0.2958763041286958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121415610041, "tokens_per_second": 151747.01386022512, "data_wait_s": 1.4760000340174884e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:10c33cd77f6f24577791df3b59d7732f855026b9ac1dc98667fb8510b8062f4c
-size 626772
+oid sha256:45447a265039cba183eb8172753e1547c755e660260e70e69af291dc2287fc2b
+size 628890