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": 76689.73514938472,
-      "required_pages_per_second_upper_estimate": 0.005305573822335977
+      "required_token_positions_per_second": 77497.94562605202,
+      "required_pages_per_second_upper_estimate": 0.005361487698418494
     },
     {
       "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": 9586.21689367309,
-      "required_pages_per_second_upper_estimate": 0.004013548830408618
+      "required_token_positions_per_second": 9687.243203256503,
+      "required_pages_per_second_upper_estimate": 0.004055846436561951
     },
     {
       "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": 19172.43378734618,
-      "required_pages_per_second_upper_estimate": 0.003993112346824364
+      "required_token_positions_per_second": 19374.486406513006,
+      "required_pages_per_second_upper_estimate": 0.0040351945789109216
     },
     {
       "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": 28758.65068101927,
-      "required_pages_per_second_upper_estimate": 0.0017240371964100977
+      "required_token_positions_per_second": 29061.72960976951,
+      "required_pages_per_second_upper_estimate": 0.001742206315413946
     },
     {
       "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": 28758.65068101927,
-      "required_pages_per_second_upper_estimate": 0.002235858594513405
+      "required_token_positions_per_second": 29061.72960976951,
+      "required_pages_per_second_upper_estimate": 0.0022594216481204156
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28758.65068101927,
-      "required_pages_per_second_upper_estimate": 0.3688850923028087
+      "required_token_positions_per_second": 29061.72960976951,
+      "required_pages_per_second_upper_estimate": 0.3727726633800171
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12374689964809704,
-    "producer_p95_seconds": 0.2969493060008972,
-    "gpu_update_seconds": 0.9613802923739986,
-    "producer_batches_per_second": 8.081010537183005,
-    "gpu_batches_per_second": 1.0401711039141808,
-    "producer_headroom": 7.76892427291436,
+    "producer_batch_seconds": 0.10669757364041743,
+    "producer_p95_seconds": 0.31590640200010967,
+    "gpu_update_seconds": 0.9513542508050084,
+    "producer_batches_per_second": 9.372284353626542,
+    "gpu_batches_per_second": 1.0511331600755756,
+    "producer_headroom": 8.91636255957588,
     "sustainable": true,
-    "buffer_seconds": 60.566958419561914,
+    "buffer_seconds": 60.886672051520534,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353411.1862394783,
-    "upload_bytes_per_second": 5844464.712245397,
+    "generation_bytes_per_second": 353106.2999527438,
+    "upload_bytes_per_second": 5561965.2425649725,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.83381357563269,
+    "estimated_full_shards_per_hour": 75.76839207588897,
     "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": 1.3237921879990608,
-      "bytes": 13505764,
-      "pages": 42,
-      "shards": 5
-    },
-    {
-      "seconds": 1.3861956760010798,
-      "bytes": 10089467,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.1900551170001563,
       "bytes": 2519802,
@@ -562,6 +550,18 @@
       "bytes": 10546134,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1281011890023365,
+      "bytes": 2729862,
+      "pages": 38,
+      "shards": 2
+    },
+    {
+      "seconds": 2.671323365997523,
+      "bytes": 5596441,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -348,3 +348,4 @@
 {"step": 519800, "loss": 0.6011606454849243, "lr": 0.0003, "grad_norm": 0.16392495698426476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76081282985, "tokens_per_second": 153256.86754452638, "data_wait_s": 1.2156997399870306e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 519900, "loss": 0.5767462253570557, "lr": 0.0003, "grad_norm": 0.16580639855739068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76096065803, "tokens_per_second": 154086.80522905567, "data_wait_s": 1.3766002666670829e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 520000, "loss": 0.6007126569747925, "lr": 0.0003, "grad_norm": 0.15556849611963117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76110796379, "tokens_per_second": 153679.85633253094, "data_wait_s": 1.1628995707724243e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 520100, "loss": 0.5963959693908691, "lr": 0.0003, "grad_norm": 0.15683531833211747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76125428033, "tokens_per_second": 133742.67439068158, "data_wait_s": 1.5700003132224083e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:58f93f5fccf0708be456267b065e13fe0e5ffea840d65e7e21201adf6d435307
-size 746946
+oid sha256:b2b7317a33804b45eb40bb38eb37071afd584a093742d0aa4c42c60553b81892
+size 749064