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": 75094.83411572932,
-      "required_pages_per_second_upper_estimate": 0.006509029895034397
+      "required_token_positions_per_second": 75840.24944039292,
+      "required_pages_per_second_upper_estimate": 0.006573640606138369
     },
     {
       "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": 9386.854264466165,
-      "required_pages_per_second_upper_estimate": 0.0037445565120736257
+      "required_token_positions_per_second": 9480.031180049114,
+      "required_pages_per_second_upper_estimate": 0.003781726176818699
     },
     {
       "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": 18773.70852893233,
-      "required_pages_per_second_upper_estimate": 0.004342967062183151
+      "required_token_positions_per_second": 18960.06236009823,
+      "required_pages_per_second_upper_estimate": 0.004386076741307969
     },
     {
       "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": 28160.56279339849,
-      "required_pages_per_second_upper_estimate": 0.0017148305270634852
+      "required_token_positions_per_second": 28440.093540147343,
+      "required_pages_per_second_upper_estimate": 0.0017318524829560058
     },
     {
       "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": 28160.56279339849,
-      "required_pages_per_second_upper_estimate": 0.002167878077262869
+      "required_token_positions_per_second": 28440.093540147343,
+      "required_pages_per_second_upper_estimate": 0.0021893971279382278
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28160.56279339849,
-      "required_pages_per_second_upper_estimate": 0.3496599425530935
+      "required_token_positions_per_second": 28440.093540147343,
+      "required_pages_per_second_upper_estimate": 0.3531307788984857
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11522391932066967,
-    "producer_p95_seconds": 0.3876982180008781,
-    "gpu_update_seconds": 0.9817985600231453,
-    "producer_batches_per_second": 8.678753559987722,
-    "gpu_batches_per_second": 1.0185388741825265,
-    "producer_headroom": 8.520787747991692,
+    "producer_batch_seconds": 0.12393196518780769,
+    "producer_p95_seconds": 0.3823251249996247,
+    "gpu_update_seconds": 0.9721487013033489,
+    "producer_batches_per_second": 8.068943298725154,
+    "gpu_batches_per_second": 1.0286492165851904,
+    "producer_headroom": 7.844212748746019,
     "sustainable": true,
-    "buffer_seconds": 60.87151072143501,
+    "buffer_seconds": 60.273219480807626,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 309971.9510090193,
-    "upload_bytes_per_second": 4059970.7144700503,
+    "generation_bytes_per_second": 311243.5752338271,
+    "upload_bytes_per_second": 4012939.2772291503,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.51276490881857,
+    "estimated_full_shards_per_hour": 66.7856258655654,
     "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": 1.235029557999951,
-      "bytes": 12065385,
-      "pages": 45,
-      "shards": 3
-    },
     {
       "seconds": 0.9373754530024598,
       "bytes": 2569463,
@@ -562,6 +556,12 @@
       "bytes": 4963940,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.0047582629995304,
+      "bytes": 6204635,
+      "pages": 42,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -336,3 +336,4 @@
 {"step": 348600, "loss": 0.5889758747071028, "lr": 0.0003, "grad_norm": 0.14416036690160747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51026133629, "tokens_per_second": 147350.11049295284, "data_wait_s": 1.1524003639351577e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 348700, "loss": 0.6345112770795822, "lr": 0.0003, "grad_norm": 0.1687719276716483, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51040607783, "tokens_per_second": 147103.8436790411, "data_wait_s": 2.0815998141188174e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 348800, "loss": 0.6054599732160568, "lr": 0.0003, "grad_norm": 0.1704089607327295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51055129564, "tokens_per_second": 148341.8375666809, "data_wait_s": 1.49589977809228e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 348900, "loss": 0.5869564954191446, "lr": 0.0003, "grad_norm": 0.15196084157394468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51069817667, "tokens_per_second": 149364.34447103136, "data_wait_s": 1.632799830986187e-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 sha256:7835b62f254de27ed55dc76dfa387601b41669b1776b139d7d34e601f9678b7e
-size 720747
+oid sha256:847c6b876081033fae0f4e5ce60b615ff4c1a7d83a1b4519f44b5833253627e1
+size 722865