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": 74416.55886809019,
-      "required_pages_per_second_upper_estimate": 0.006450238715641935
+      "required_token_positions_per_second": 72608.12507106553,
+      "required_pages_per_second_upper_estimate": 0.006293488257549393
     },
     {
       "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": 9302.069858511273,
-      "required_pages_per_second_upper_estimate": 0.003710734744898386
+      "required_token_positions_per_second": 9076.015633883191,
+      "required_pages_per_second_upper_estimate": 0.0036205583348823964
     },
     {
       "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": 18604.139717022546,
-      "required_pages_per_second_upper_estimate": 0.004303740301856985
+      "required_token_positions_per_second": 18152.031267766382,
+      "required_pages_per_second_upper_estimate": 0.004199152968958519
     },
     {
       "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": 27906.20957553382,
-      "required_pages_per_second_upper_estimate": 0.001699341750583726
+      "required_token_positions_per_second": 27228.046901649577,
+      "required_pages_per_second_upper_estimate": 0.0016580452017887454
     },
     {
       "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": 27906.20957553382,
-      "required_pages_per_second_upper_estimate": 0.0021482972624568756
+      "required_token_positions_per_second": 27228.046901649577,
+      "required_pages_per_second_upper_estimate": 0.0020960904225468344
     },
     {
       "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": 27906.20957553382,
-      "required_pages_per_second_upper_estimate": 0.34650172685267416
+      "required_token_positions_per_second": 27228.046901649577,
+      "required_pages_per_second_upper_estimate": 0.33808121610749814
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.14553018894537217,
-    "producer_p95_seconds": 0.36363843100116355,
-    "gpu_update_seconds": 0.9907472358496082,
-    "producer_batches_per_second": 6.871426521512805,
-    "gpu_batches_per_second": 1.0093391773558238,
-    "producer_headroom": 6.8078468325325,
+    "producer_batch_seconds": 0.1474725581561529,
+    "producer_p95_seconds": 0.36600135800108546,
+    "gpu_update_seconds": 1.0154235483678773,
+    "producer_batches_per_second": 6.780922583177402,
+    "gpu_batches_per_second": 0.9848107241626728,
+    "producer_headroom": 6.885508470617871,
     "sustainable": true,
-    "buffer_seconds": 60.4355813868261,
+    "buffer_seconds": 60.92541290207264,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 257926.16875206766,
-    "upload_bytes_per_second": 5235190.093287172,
+    "generation_bytes_per_second": 292542.9964676456,
+    "upload_bytes_per_second": 5268048.961934955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.34495160028002,
+    "estimated_full_shards_per_hour": 62.772916989536526,
     "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": 1.269822567999654,
-      "bytes": 1251620,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 0.9414359830007015,
       "bytes": 5894779,
@@ -562,6 +556,12 @@
       "bytes": 6082322,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.977721251998446,
+      "bytes": 2308590,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -362,3 +362,4 @@
 {"step": 273300, "loss": 0.6495624426752329, "lr": 0.0003, "grad_norm": 0.16367195483570454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40002486610, "tokens_per_second": 118874.44948911833, "data_wait_s": 1.1262993211857975e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 273400, "loss": 0.5447521097958088, "lr": 0.0003, "grad_norm": 0.1448215740828469, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40017066899, "tokens_per_second": 148404.4334698735, "data_wait_s": 9.012997907120734e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 273500, "loss": 0.6166048962622881, "lr": 0.0003, "grad_norm": 0.1644891949652567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40031603668, "tokens_per_second": 148641.05553318723, "data_wait_s": 1.8188002286478877e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 273600, "loss": 0.5984455365687609, "lr": 0.0003, "grad_norm": 0.1406793601643985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40046282027, "tokens_per_second": 149426.92946310755, "data_wait_s": 9.884999599307775e-06, "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:fd076230c815300ee372c20dfe334d8efbdd974a9a0cdf1df36ec1b8ec78d283
-size 776598
+oid sha256:2a3f06035d1c0af3de2d790bceb1808df4a9e175edb77071f815dc34cd7aaf9e
+size 778716