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": 75446.98432646661,
-      "required_pages_per_second_upper_estimate": 0.005219597436827313
+      "required_token_positions_per_second": 76237.62216600477,
+      "required_pages_per_second_upper_estimate": 0.0052742955970991
     },
     {
       "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": 9430.873040808327,
-      "required_pages_per_second_upper_estimate": 0.003948509603162671
+      "required_token_positions_per_second": 9529.702770750597,
+      "required_pages_per_second_upper_estimate": 0.003989887547290056
     },
     {
       "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": 18861.746081616653,
-      "required_pages_per_second_upper_estimate": 0.0039284042911066856
+      "required_token_positions_per_second": 19059.405541501194,
+      "required_pages_per_second_upper_estimate": 0.003969571543970144
     },
     {
       "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": 28292.619122424978,
-      "required_pages_per_second_upper_estimate": 0.0016960993160613576
+      "required_token_positions_per_second": 28589.10831225179,
+      "required_pages_per_second_upper_estimate": 0.0017138733902786949
     },
     {
       "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": 28292.619122424978,
-      "required_pages_per_second_upper_estimate": 0.0021996266906888894
+      "required_token_positions_per_second": 28589.10831225179,
+      "required_pages_per_second_upper_estimate": 0.0022226774210798014
     },
     {
       "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": 28292.619122424978,
-      "required_pages_per_second_upper_estimate": 0.36290733985486306
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28589.10831225179,
+      "required_pages_per_second_upper_estimate": 0.366710384836672
     }
   ],
   "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": 0.31690019344537745,
-    "producer_p95_seconds": 0.29973960800271016,
-    "gpu_update_seconds": 0.9772159968776434,
-    "producer_batches_per_second": 3.155567654054983,
-    "gpu_batches_per_second": 1.0233152171015978,
-    "producer_headroom": 3.0836711907721868,
+    "producer_batch_seconds": 0.0972974863358047,
+    "producer_p95_seconds": 0.288149070001964,
+    "gpu_update_seconds": 0.9670815786916838,
+    "producer_batches_per_second": 10.2777578091656,
+    "gpu_batches_per_second": 1.0340389291179033,
+    "producer_headroom": 9.939430247498649,
     "sustainable": true,
-    "buffer_seconds": 60.58739180641389,
+    "buffer_seconds": 60.92613945757608,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 392385.99433768494,
-    "upload_bytes_per_second": 3257031.791403977,
+    "generation_bytes_per_second": 408056.40612637537,
+    "upload_bytes_per_second": 3222862.9991209614,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.19690010641013,
+    "estimated_full_shards_per_hour": 87.55940568774649,
     "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": 1.6191135740009486,
-      "bytes": 10591273,
-      "pages": 42,
-      "shards": 3
-    },
     {
       "seconds": 1.5766091279983812,
       "bytes": 7345474,
@@ -562,6 +556,12 @@
       "bytes": 11600439,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.7552413789999264,
+      "bytes": 6467958,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -271,3 +271,4 @@
 {"step": 827100, "loss": 0.5386779308319092, "lr": 3e-05, "grad_norm": 0.2909685545523433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121064265463, "tokens_per_second": 152396.25845027762, "data_wait_s": 9.840998245636001e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 827200, "loss": 0.481920063495636, "lr": 3e-05, "grad_norm": 0.3127427065558501, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121078877060, "tokens_per_second": 149922.61414437514, "data_wait_s": 8.499999239575118e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 827300, "loss": 0.5415542125701904, "lr": 3e-05, "grad_norm": 0.29402953904389895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121093286690, "tokens_per_second": 149999.84414501625, "data_wait_s": 8.249000529758632e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 827400, "loss": 0.5174568891525269, "lr": 3e-05, "grad_norm": 0.2933049902606767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121107699509, "tokens_per_second": 148877.0362265517, "data_wait_s": 7.4979980126954615e-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 sha256:1765a02eecd338d03b76e81bba1b0356a73a61e06a71f7b27b259f26d5a1ac64
-size 582294
+oid sha256:ec21a4ebe222176d6d7b698f7b9062c16dbde69039ec72653ebc279ca253df90
+size 584412