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": 75370.20022632885,
-      "required_pages_per_second_upper_estimate": 0.006532897931592137
+      "required_token_positions_per_second": 74971.46880314293,
+      "required_pages_per_second_upper_estimate": 0.006498336902405934
     },
     {
       "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": 9421.275028291106,
-      "required_pages_per_second_upper_estimate": 0.0037582874693996753
+      "required_token_positions_per_second": 9371.433600392866,
+      "required_pages_per_second_upper_estimate": 0.0037384049786153128
     },
     {
       "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": 18842.55005658221,
-      "required_pages_per_second_upper_estimate": 0.004358892338035445
+      "required_token_positions_per_second": 18742.86720078573,
+      "required_pages_per_second_upper_estimate": 0.0043358324637052725
     },
     {
       "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": 28263.82508487332,
-      "required_pages_per_second_upper_estimate": 0.0017211186588389294
+      "required_token_positions_per_second": 28114.300801178597,
+      "required_pages_per_second_upper_estimate": 0.0017120134144552085
     },
     {
       "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": 28263.82508487332,
-      "required_pages_per_second_upper_estimate": 0.0021758274943089193
+      "required_token_positions_per_second": 28114.300801178597,
+      "required_pages_per_second_upper_estimate": 0.002164316701040391
     },
     {
       "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": 28263.82508487332,
-      "required_pages_per_second_upper_estimate": 0.3509421146165529
+      "required_token_positions_per_second": 28114.300801178597,
+      "required_pages_per_second_upper_estimate": 0.3490855234386505
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.10945598122637534,
-    "producer_p95_seconds": 0.3148644939974474,
-    "gpu_update_seconds": 0.9782115448625917,
-    "producer_batches_per_second": 9.136092781734913,
-    "gpu_batches_per_second": 1.022273766090615,
-    "producer_headroom": 8.937031434028881,
+    "producer_batch_seconds": 0.13281460116414223,
+    "producer_p95_seconds": 0.35745015499924193,
+    "gpu_update_seconds": 0.983414106419497,
+    "producer_batches_per_second": 7.529292647305586,
+    "gpu_batches_per_second": 1.0168656250426287,
+    "producer_headroom": 7.404412600720912,
     "sustainable": true,
-    "buffer_seconds": 60.64911578148068,
+    "buffer_seconds": 60.97167459800882,
     "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": 290024.70269958174,
-    "upload_bytes_per_second": 3645490.806645077,
+    "generation_bytes_per_second": 320003.39240656176,
+    "upload_bytes_per_second": 3609103.746687113,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.23254976978864,
+    "estimated_full_shards_per_hour": 68.66527871272697,
     "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": 1.126622554998903,
-      "bytes": 6643361,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 0.8856843470020976,
       "bytes": 5828359,
@@ -562,6 +556,12 @@
       "bytes": 6233089,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.140030457998364,
+      "bytes": 2525214,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -315,3 +315,4 @@
 {"step": 71500, "loss": 0.5960981119424105, "lr": 0.0003, "grad_norm": 0.13603921606396208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10464120284, "tokens_per_second": 147566.92890491622, "data_wait_s": 8.75100158737041e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 71600, "loss": 0.5503122769296169, "lr": 0.0003, "grad_norm": 0.13911065791274513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10478668955, "tokens_per_second": 149021.7005775473, "data_wait_s": 1.588200029800646e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 71700, "loss": 0.5870297811925411, "lr": 0.0003, "grad_norm": 0.12916020390820285, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10493302798, "tokens_per_second": 149436.38231045956, "data_wait_s": 9.063998732017353e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 71800, "loss": 0.576345793902874, "lr": 0.0003, "grad_norm": 0.14564972279567823, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10508007674, "tokens_per_second": 150312.12327819463, "data_wait_s": 8.13999940874055e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:978ab0b4d8c7eaeaf2eb6ffb30bd8f790fdbc56aa90f55344a9552e40f882c1d
-size 676269
+oid sha256:3cff02c4b6c227bd61518b13f1b9860bedbfbd863991df5819f2105eb7c1d25e
+size 678387