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": 76093.30843885786,
-      "required_pages_per_second_upper_estimate": 0.005264311638601096
+      "required_token_positions_per_second": 76711.50807713918,
+      "required_pages_per_second_upper_estimate": 0.005307080123998184
     },
     {
       "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": 9511.663554857232,
-      "required_pages_per_second_upper_estimate": 0.003982334904297168
+      "required_token_positions_per_second": 9588.938509642398,
+      "required_pages_per_second_upper_estimate": 0.0040146883141811635
     },
     {
       "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": 19023.327109714464,
-      "required_pages_per_second_upper_estimate": 0.003962057358081196
+      "required_token_positions_per_second": 19177.877019284795,
+      "required_pages_per_second_upper_estimate": 0.003994246028489499
     },
     {
       "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": 28534.9906645717,
-      "required_pages_per_second_upper_estimate": 0.001710629119933137
+      "required_token_positions_per_second": 28766.815528927193,
+      "required_pages_per_second_upper_estimate": 0.0017245266665751769
     },
     {
       "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": 28534.9906645717,
-      "required_pages_per_second_upper_estimate": 0.00221847001201105
+      "required_token_positions_per_second": 28766.815528927193,
+      "required_pages_per_second_upper_estimate": 0.0022364933755248755
     },
     {
       "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": 28534.9906645717,
-      "required_pages_per_second_upper_estimate": 0.3660162217592347
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28766.815528927193,
+      "required_pages_per_second_upper_estimate": 0.3689898222050409
     }
   ],
   "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": 0.1038720019139987,
-    "producer_p95_seconds": 0.3085875719989417,
-    "gpu_update_seconds": 0.9689156840807044,
-    "producer_batches_per_second": 9.627233340779881,
-    "gpu_batches_per_second": 1.0320815489211408,
-    "producer_headroom": 9.327977378186304,
+    "producer_batch_seconds": 0.10546405768036493,
+    "producer_p95_seconds": 0.31745230999513296,
+    "gpu_update_seconds": 0.9611074250536303,
+    "producer_batches_per_second": 9.481903332704578,
+    "gpu_batches_per_second": 1.0404664181469614,
+    "producer_headroom": 9.113127696703133,
     "sustainable": true,
-    "buffer_seconds": 60.07277241300367,
+    "buffer_seconds": 60.54976777837871,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 265336.07756232255,
-    "upload_bytes_per_second": 3189645.75749903,
+    "generation_bytes_per_second": 252940.97603613717,
+    "upload_bytes_per_second": 3145185.400648082,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.93494553711183,
+    "estimated_full_shards_per_hour": 54.27524529278838,
     "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": 1.5976459840021562,
-      "bytes": 11304870,
-      "pages": 44,
-      "shards": 3
-    },
     {
       "seconds": 1.606354307001311,
       "bytes": 7330991,
@@ -562,6 +556,12 @@
       "bytes": 11276737,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.8010207819970674,
+      "bytes": 5858104,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -321,3 +321,4 @@
 {"step": 832100, "loss": 0.5186933279037476, "lr": 3e-05, "grad_norm": 0.28658392748029904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121796373298, "tokens_per_second": 137708.14166155068, "data_wait_s": 1.0030002158600837e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 832200, "loss": 0.5327817797660828, "lr": 3e-05, "grad_norm": 0.32483344683531323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121810915468, "tokens_per_second": 150440.65964355343, "data_wait_s": 1.3541997759602964e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 832300, "loss": 0.5248480439186096, "lr": 3e-05, "grad_norm": 0.275588656874606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121825346988, "tokens_per_second": 149627.26603466485, "data_wait_s": 1.3630000466946512e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 832400, "loss": 0.5739641785621643, "lr": 3e-05, "grad_norm": 0.284707157600611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121839877898, "tokens_per_second": 116365.84335738016, "data_wait_s": 8.529998012818396e-06, "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:e78de16f897f66f8326ca26640cf892eb58541f5beb150b0fc75cc2447240f88
-size 688977
+oid sha256:e86343d0a42c55ac4cf7c59ec84ed084a566febbf1d8f3d2083e98dc077eb0a0
+size 691095