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": 75517.6239183727,
-      "required_pages_per_second_upper_estimate": 0.005224484447701812
+      "required_token_positions_per_second": 76322.36720738787,
+      "required_pages_per_second_upper_estimate": 0.00528015845569757
     },
     {
       "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": 9439.702989796588,
-      "required_pages_per_second_upper_estimate": 0.0039522065184137535
+      "required_token_positions_per_second": 9540.295900923484,
+      "required_pages_per_second_upper_estimate": 0.003994322669683731
     },
     {
       "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": 18879.405979593175,
-      "required_pages_per_second_upper_estimate": 0.0039320823821323665
+      "required_token_positions_per_second": 19080.591801846967,
+      "required_pages_per_second_upper_estimate": 0.003973984083280911
     },
     {
       "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": 28319.108969389763,
-      "required_pages_per_second_upper_estimate": 0.0016976873419392421
+      "required_token_positions_per_second": 28620.88770277045,
+      "required_pages_per_second_upper_estimate": 0.0017157785162159691
     },
     {
       "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": 28319.108969389763,
-      "required_pages_per_second_upper_estimate": 0.0022016861597738786
+      "required_token_positions_per_second": 28620.88770277045,
+      "required_pages_per_second_upper_estimate": 0.0022251481289098612
     },
     {
       "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": 28319.108969389763,
-      "required_pages_per_second_upper_estimate": 0.36324712316914565
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28620.88770277045,
+      "required_pages_per_second_upper_estimate": 0.3671180167361944
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.15657385643768862,
-    "producer_p95_seconds": 0.30418533899864997,
-    "gpu_update_seconds": 0.97630190377405,
-    "producer_batches_per_second": 6.386762277890038,
-    "gpu_batches_per_second": 1.0242733278859144,
-    "producer_headroom": 6.235408170856332,
+    "producer_batch_seconds": 0.0992931515938551,
+    "producer_p95_seconds": 0.30434061599953566,
+    "gpu_update_seconds": 0.9660077733131849,
+    "producer_batches_per_second": 10.071188032084645,
+    "gpu_batches_per_second": 1.0351883573050655,
+    "producer_headroom": 9.728845925492486,
     "sustainable": true,
-    "buffer_seconds": 60.5307180339911,
+    "buffer_seconds": 60.858489718730645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 431498.2319016139,
-    "upload_bytes_per_second": 5735935.499491359,
+    "generation_bytes_per_second": 415962.73691658623,
+    "upload_bytes_per_second": 5702473.091301636,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.58947580133736,
+    "estimated_full_shards_per_hour": 89.25592022536459,
     "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": 1.3151332330016885,
-      "bytes": 10337557,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.029109121998772,
       "bytes": 7643162,
@@ -562,6 +556,12 @@
       "bytes": 5868379,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 0.9594618119990628,
+      "bytes": 5817766,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 586500, "loss": 0.5583824515342712, "lr": 0.0003, "grad_norm": 0.17990932988075062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85846333053, "tokens_per_second": 151401.49081931167, "data_wait_s": 9.75199873209931e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 586600, "loss": 0.5461352467536926, "lr": 0.0003, "grad_norm": 0.16201072949906475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85860814427, "tokens_per_second": 149998.77965041617, "data_wait_s": 2.0488001609919593e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 586700, "loss": 0.5536483526229858, "lr": 0.0003, "grad_norm": 0.14367783967544867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85875449788, "tokens_per_second": 152037.35700774894, "data_wait_s": 8.17299951449968e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 586800, "loss": 0.5427125692367554, "lr": 0.0003, "grad_norm": 0.13986188715868894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85890049418, "tokens_per_second": 152087.50209607932, "data_wait_s": 1.151100150309503e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b19ac0c562d48a8d5953188cb7ba1f5d7a702bc61452638ab080209812e15824
-size 631791
+oid sha256:ccec6908f6dea90b860a7b2751d4a37942ec2e234149ffde79f24cfd5f4f532c
+size 633909