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": 76160.8000497871,
-      "required_pages_per_second_upper_estimate": 0.005268980864847294
+      "required_token_positions_per_second": 76337.35860512854,
+      "required_pages_per_second_upper_estimate": 0.005281195595378105
     },
     {
       "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": 9520.100006223387,
-      "required_pages_per_second_upper_estimate": 0.00398586707031104
+      "required_token_positions_per_second": 9542.169825641067,
+      "required_pages_per_second_upper_estimate": 0.003995107242830986
     },
     {
       "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": 19040.200012446774,
-      "required_pages_per_second_upper_estimate": 0.0039655715387519695
+      "required_token_positions_per_second": 19084.339651282135,
+      "required_pages_per_second_upper_estimate": 0.00397476466148082
     },
     {
       "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": 28560.300018670165,
-      "required_pages_per_second_upper_estimate": 0.001712146377066194
+      "required_token_positions_per_second": 28626.509476923202,
+      "required_pages_per_second_upper_estimate": 0.0017161155329924744
     },
     {
       "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": 28560.300018670165,
-      "required_pages_per_second_upper_estimate": 0.002220437702968122
+      "required_token_positions_per_second": 28626.509476923202,
+      "required_pages_per_second_upper_estimate": 0.0022255851971226667
     },
     {
       "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": 28560.300018670165,
-      "required_pages_per_second_upper_estimate": 0.3663408629785427
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28626.509476923202,
+      "required_pages_per_second_upper_estimate": 0.36719012681883506
     }
   ],
   "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": 0.06895367942969699,
-    "producer_p95_seconds": 0.22242711199942278,
-    "gpu_update_seconds": 0.9680570575913494,
-    "producer_batches_per_second": 14.502489327194914,
-    "gpu_batches_per_second": 1.0329969624808364,
-    "producer_headroom": 14.039237145834257,
+    "producer_batch_seconds": 0.08480372392972413,
+    "producer_p95_seconds": 0.234814481000285,
+    "gpu_update_seconds": 0.9658180653246596,
+    "producer_batches_per_second": 11.791934996023153,
+    "gpu_batches_per_second": 1.0353916911502894,
+    "producer_headroom": 11.38886384429323,
     "sustainable": true,
-    "buffer_seconds": 60.019537570663665,
+    "buffer_seconds": 60.846538115453555,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356164.7019734799,
-    "upload_bytes_per_second": 5004250.596844194,
+    "generation_bytes_per_second": 410642.5200900546,
+    "upload_bytes_per_second": 4948377.641518796,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.42465395358369,
+    "estimated_full_shards_per_hour": 88.11432554270009,
     "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": 1.3861621640007797,
-      "bytes": 7683491,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.2777914989992496,
       "bytes": 16450745,
@@ -562,6 +556,12 @@
       "bytes": 7984438,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.001051560999258,
+      "bytes": 939443,
+      "pages": 30,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -135,3 +135,4 @@
 {"step": 733500, "loss": 0.5229381918907166, "lr": 5e-05, "grad_norm": 0.23257659318026108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107364245598, "tokens_per_second": 151244.1606306, "data_wait_s": 1.8706999981077388e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 733600, "loss": 0.5356453061103821, "lr": 5e-05, "grad_norm": 0.2400567351541834, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107379064228, "tokens_per_second": 151748.4798672924, "data_wait_s": 1.813200105971191e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 733700, "loss": 0.5465325117111206, "lr": 5e-05, "grad_norm": 0.22246921809149486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107393768791, "tokens_per_second": 151113.96318122078, "data_wait_s": 1.57409995154012e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 733800, "loss": 0.5684446692466736, "lr": 5e-05, "grad_norm": 0.2369546057857919, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 107408502476, "tokens_per_second": 150686.9003358025, "data_wait_s": 1.927700031956192e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cb928c5eee659095a0485506f7f449cfe12486b9f554e709782d016df2e8871a
-size 291897
+oid sha256:c4cad716dd0fd7ab5c934e913cac4246ae0b7e8c3e8244eb896048074fb5916a
+size 294015