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": 75250.54825080973,
-      "required_pages_per_second_upper_estimate": 0.0065225268016093005
+      "required_token_positions_per_second": 75540.15889024008,
+      "required_pages_per_second_upper_estimate": 0.006547629517823668
     },
     {
       "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": 9406.318531351217,
-      "required_pages_per_second_upper_estimate": 0.0037523210991507964
+      "required_token_positions_per_second": 9442.51986128001,
+      "required_pages_per_second_upper_estimate": 0.0037667623509175085
     },
     {
       "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": 18812.637062702433,
-      "required_pages_per_second_upper_estimate": 0.00435197249335207
+      "required_token_positions_per_second": 18885.03972256002,
+      "required_pages_per_second_upper_estimate": 0.00436872157446683
     },
     {
       "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": 28218.955594053652,
-      "required_pages_per_second_upper_estimate": 0.0017183863422600364
+      "required_token_positions_per_second": 28327.55958384003,
+      "required_pages_per_second_upper_estimate": 0.001724999755436929
     },
     {
       "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": 28218.955594053652,
-      "required_pages_per_second_upper_estimate": 0.0021723733166989214
+      "required_token_positions_per_second": 28327.55958384003,
+      "required_pages_per_second_upper_estimate": 0.002180733952467762
     },
     {
       "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": 28218.955594053652,
-      "required_pages_per_second_upper_estimate": 0.3503849857090983
+      "required_token_positions_per_second": 28327.55958384003,
+      "required_pages_per_second_upper_estimate": 0.35173348378807295
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09879406045315164,
-    "producer_p95_seconds": 0.19274397499975748,
-    "gpu_update_seconds": 0.9797669480661444,
-    "producer_batches_per_second": 10.122065996813665,
-    "gpu_batches_per_second": 1.0206508823080749,
-    "producer_headroom": 9.917265709822221,
+    "producer_batch_seconds": 0.18137553756248082,
+    "producer_p95_seconds": 0.18334560500079533,
+    "gpu_update_seconds": 0.9760106555656952,
+    "producer_batches_per_second": 5.513422666799909,
+    "gpu_batches_per_second": 1.0245789780034733,
+    "producer_headroom": 5.381159271434143,
     "sustainable": true,
-    "buffer_seconds": 60.745550780100956,
+    "buffer_seconds": 60.512660645073105,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 404740.10877153405,
-    "upload_bytes_per_second": 5355263.17093906,
+    "generation_bytes_per_second": 393845.4131474396,
+    "upload_bytes_per_second": 5341890.859447985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.8478054748489,
+    "estimated_full_shards_per_hour": 84.51005740945236,
     "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": [
@@ -478,6 +478,12 @@
       "bytes": 7586010,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0743332829997598,
+      "bytes": 4951926,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -59,3 +59,4 @@
 {"step": 320900, "loss": 0.5831205900758505, "lr": 0.0003, "grad_norm": 0.15360701551341333, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46970679323, "tokens_per_second": 150797.79150006865, "data_wait_s": 1.4571000065188855e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 321000, "loss": 0.6045909468084574, "lr": 0.0003, "grad_norm": 0.1401154045649198, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46985391259, "tokens_per_second": 149704.20951458134, "data_wait_s": 9.342999874206726e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 321100, "loss": 0.6374973971396685, "lr": 0.0003, "grad_norm": 0.19649614827835046, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47000203249, "tokens_per_second": 150947.57448954, "data_wait_s": 1.3704000593861565e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 321200, "loss": 0.6152104809880257, "lr": 0.0003, "grad_norm": 0.15699181575060958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47014798356, "tokens_per_second": 148378.56027400636, "data_wait_s": 1.3146000128472224e-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 sha256:9459522ba6c4b38ddbdfb472a78201355f3e2733fc79825bc1694a1aa607f6ab
-size 130146
+oid sha256:464c8fedb43c25f290bd4679715fc593f810fe2f7d21728f05fb5fe6c532144f
+size 132264