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": 76003.15056315946,
-      "required_pages_per_second_upper_estimate": 0.006587760462590653
+      "required_token_positions_per_second": 75509.24353657718,
+      "required_pages_per_second_upper_estimate": 0.006544949853322415
     },
     {
       "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": 9500.393820394933,
-      "required_pages_per_second_upper_estimate": 0.0037898491385012496
+      "required_token_positions_per_second": 9438.655442072148,
+      "required_pages_per_second_upper_estimate": 0.003765220776317276
     },
     {
       "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": 19000.787640789866,
-      "required_pages_per_second_upper_estimate": 0.004395497818255519
+      "required_token_positions_per_second": 18877.310884144295,
+      "required_pages_per_second_upper_estimate": 0.004366933643722262
     },
     {
       "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": 28501.1814611848,
-      "required_pages_per_second_upper_estimate": 0.0017355724168436348
+      "required_token_positions_per_second": 28315.966326216443,
+      "required_pages_per_second_upper_estimate": 0.0017242937868727744
     },
     {
       "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": 28501.1814611848,
-      "required_pages_per_second_upper_estimate": 0.0021940998452018795
+      "required_token_positions_per_second": 28315.966326216443,
+      "required_pages_per_second_upper_estimate": 0.002179841471403707
     },
     {
       "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": 28501.1814611848,
-      "required_pages_per_second_upper_estimate": 0.35388928642965095
+      "required_token_positions_per_second": 28315.966326216443,
+      "required_pages_per_second_upper_estimate": 0.3515895343285253
     }
   ],
   "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": 0.07956427864058924,
-    "producer_p95_seconds": 0.3020958100023563,
-    "gpu_update_seconds": 0.9700650493262277,
-    "producer_batches_per_second": 12.568454299915642,
-    "gpu_batches_per_second": 1.0308587044699364,
-    "producer_headroom": 12.192218240402106,
+    "producer_batch_seconds": 0.13886678171857625,
+    "producer_p95_seconds": 0.34607905899974867,
+    "gpu_update_seconds": 0.9764102584908781,
+    "producer_batches_per_second": 7.201146218154415,
+    "gpu_batches_per_second": 1.0241596616831758,
+    "producer_headroom": 7.031273040298762,
     "sustainable": true,
-    "buffer_seconds": 60.14403305822612,
+    "buffer_seconds": 60.53743602643444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300932.19316467526,
-    "upload_bytes_per_second": 4463118.677455215,
+    "generation_bytes_per_second": 368890.5155984655,
+    "upload_bytes_per_second": 4426820.98066969,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.57304331021493,
+    "estimated_full_shards_per_hour": 79.15531731572604,
     "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": 1.392480271999375,
-      "bytes": 8010893,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.3167049000003317,
       "bytes": 10407908,
@@ -562,6 +556,12 @@
       "bytes": 6494530,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9434360529994592,
+      "bytes": 2587569,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -271,3 +271,4 @@
 {"step": 264200, "loss": 0.6068587508052588, "lr": 0.0003, "grad_norm": 0.1476930885140546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38671799542, "tokens_per_second": 149610.1927024016, "data_wait_s": 1.0306001058779657e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 264300, "loss": 0.8233817405998707, "lr": 0.0003, "grad_norm": 0.2062728228365645, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38686528214, "tokens_per_second": 150251.655592251, "data_wait_s": 9.554998541716486e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 264400, "loss": 0.5680136904120445, "lr": 0.0003, "grad_norm": 0.14363066092926072, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38701150634, "tokens_per_second": 148686.49191587896, "data_wait_s": 1.69100021594204e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 264500, "loss": 0.6122810952365398, "lr": 0.0003, "grad_norm": 0.18052651250036417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 38715760764, "tokens_per_second": 148749.38149347337, "data_wait_s": 1.2296000932110474e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d046ae4279827107c01e8615ea9f3ec054ca7676e1a7d994245713ab76f0830
-size 582294
+oid sha256:a39c075fadc3d896b5b79f0d8247346e5a362b51c765ce2f9513cf082c389845
+size 584412