Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74933.2151958986,
-      "required_pages_per_second_upper_estimate": 0.0064950211766828785
+      "required_token_positions_per_second": 74726.12750727916,
+      "required_pages_per_second_upper_estimate": 0.006477071340692293
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9366.651899487326,
-      "required_pages_per_second_upper_estimate": 0.003736497486631293
+      "required_token_positions_per_second": 9340.765938409895,
+      "required_pages_per_second_upper_estimate": 0.0037261711897278982
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18733.30379897465,
-      "required_pages_per_second_upper_estimate": 0.004333620139006403
+      "required_token_positions_per_second": 18681.53187681979,
+      "required_pages_per_second_upper_estimate": 0.004321643615970585
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28099.95569846198,
-      "required_pages_per_second_upper_estimate": 0.001711139872962704
+      "required_token_positions_per_second": 28022.297815229686,
+      "required_pages_per_second_upper_estimate": 0.001706410915313282
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28099.95569846198,
-      "required_pages_per_second_upper_estimate": 0.00216321237532348
+      "required_token_positions_per_second": 28022.297815229686,
+      "required_pages_per_second_upper_estimate": 0.0021572340565014785
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 28099.95569846198,
-      "required_pages_per_second_upper_estimate": 0.34890740527288056
+      "required_token_positions_per_second": 28022.297815229686,
+      "required_pages_per_second_upper_estimate": 0.34794315426735145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.147282152047012,
-    "producer_p95_seconds": 0.479906843000208,
-    "gpu_update_seconds": 0.9839161419572374,
-    "producer_batches_per_second": 6.789688948059389,
-    "gpu_batches_per_second": 1.0163467772881214,
-    "producer_headroom": 6.680484554864289,
+    "producer_batch_seconds": 0.13716007199235491,
+    "producer_p95_seconds": 0.4414159650041256,
+    "gpu_update_seconds": 0.9866428578520687,
+    "producer_batches_per_second": 7.290751495491621,
+    "gpu_batches_per_second": 1.0135379707476015,
+    "producer_headroom": 7.193367891401097,
     "sustainable": true,
-    "buffer_seconds": 60.01888465939148,
+    "buffer_seconds": 60.185214328976194,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 269108.68049209326,
-    "upload_bytes_per_second": 5443538.269512217,
+    "generation_bytes_per_second": 280441.08105400676,
+    "upload_bytes_per_second": 5431960.895851179,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.74445830413912,
+    "estimated_full_shards_per_hour": 60.17612765994217,
     "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": 0.9340340310009196,
-      "bytes": 958542,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.0266037770052208,
       "bytes": 7314912,
@@ -562,6 +556,12 @@
       "bytes": 12231329,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.1164654570020502,
+      "bytes": 1063474,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -419,3 +419,4 @@
 {"step": 356900, "loss": 0.615576034411788, "lr": 0.0003, "grad_norm": 0.15941947816925933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52239903919, "tokens_per_second": 140987.3160980433, "data_wait_s": 1.2231997970957309e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357000, "loss": 0.604896292090416, "lr": 0.0003, "grad_norm": 0.15028696367232258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52254457624, "tokens_per_second": 145493.48765809817, "data_wait_s": 1.3530996511690319e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 357100, "loss": 0.5408975128084421, "lr": 0.0003, "grad_norm": 0.15872110976417061, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52269172951, "tokens_per_second": 148331.15850647166, "data_wait_s": 1.1898999218828976e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 357200, "loss": 0.6047567324712873, "lr": 0.0003, "grad_norm": 0.15845736428146975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52283949229, "tokens_per_second": 149524.7630702091, "data_wait_s": 1.6414000128861517e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac48ee77e98c0f027260357605cc20c9166c35d067fc8c47a7c31b762a800744
-size 898107
+oid sha256:fe6906dde62fb90e08768b129c0ba639e0037edcf4514126ca792143841b0934
+size 900225