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": 75534.87301804383,
-      "required_pages_per_second_upper_estimate": 0.006547171351818617
+      "required_token_positions_per_second": 73934.81293113582,
+      "required_pages_per_second_upper_estimate": 0.006408482198800645
     },
     {
       "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": 9441.859127255479,
-      "required_pages_per_second_upper_estimate": 0.003766498774236269
+      "required_token_positions_per_second": 9241.851616391978,
+      "required_pages_per_second_upper_estimate": 0.0036867127877740456
     },
     {
       "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": 18883.718254510957,
-      "required_pages_per_second_upper_estimate": 0.004368415876090722
+      "required_token_positions_per_second": 18483.703232783955,
+      "required_pages_per_second_upper_estimate": 0.004275879440838116
     },
     {
       "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": 28325.577381766434,
-      "required_pages_per_second_upper_estimate": 0.0017248790497304575
+      "required_token_positions_per_second": 27725.554849175933,
+      "required_pages_per_second_upper_estimate": 0.0016883408255706285
     },
     {
       "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": 28325.577381766434,
-      "required_pages_per_second_upper_estimate": 0.00218058135706505
+      "required_token_positions_per_second": 27725.554849175933,
+      "required_pages_per_second_upper_estimate": 0.002134389961537556
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28325.577381766434,
-      "required_pages_per_second_upper_estimate": 0.35170887147232244
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27725.554849175933,
+      "required_pages_per_second_upper_estimate": 0.34425859976378476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05088474019532896,
-    "producer_p95_seconds": 0.18023312999957852,
-    "gpu_update_seconds": 0.9760789560391239,
-    "producer_batches_per_second": 19.652257163175936,
-    "gpu_batches_per_second": 1.0245072837733809,
-    "producer_headroom": 19.18215465564516,
+    "producer_batch_seconds": 0.2630036773984372,
+    "producer_p95_seconds": 0.184500276000108,
+    "gpu_update_seconds": 0.9972027665596659,
+    "producer_batches_per_second": 3.802228204151879,
+    "gpu_batches_per_second": 1.0028050799036434,
+    "producer_headroom": 3.7915924842714435,
     "sustainable": true,
-    "buffer_seconds": 60.51689527442568,
+    "buffer_seconds": 60.829368760139616,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 521643.3788897799,
-    "upload_bytes_per_second": 5829494.409536731,
+    "generation_bytes_per_second": 490491.1786911069,
+    "upload_bytes_per_second": 5843394.615224355,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.9325258733754,
+    "estimated_full_shards_per_hour": 105.24798889684587,
     "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": [
@@ -418,6 +418,12 @@
       "bytes": 1185685,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.760915255999862,
+      "bytes": 10869507,
+      "pages": 29,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -45,3 +45,4 @@
 {"step": 86400, "loss": 0.6343935448676348, "lr": 0.0003, "grad_norm": 0.14075740858816985, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12644099192, "tokens_per_second": 147445.3847662837, "data_wait_s": 1.1063000783906318e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 86500, "loss": 0.631697554141283, "lr": 0.0003, "grad_norm": 0.13317855507861112, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12658762758, "tokens_per_second": 148322.60072593484, "data_wait_s": 1.1066000297432765e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 86600, "loss": 0.6253009773790836, "lr": 0.0003, "grad_norm": 0.12850795212147578, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12673432090, "tokens_per_second": 149181.88792785766, "data_wait_s": 1.625500044610817e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 86700, "loss": 0.6089567504823208, "lr": 0.0003, "grad_norm": 0.1368707770063171, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12687942064, "tokens_per_second": 146885.27042011803, "data_wait_s": 1.192699983221246e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f065ffaea2d398bdbd84136763c6b75d9a2aa9256c88ce68b8ad14cc1c6b5194
-size 100494
+oid sha256:a913ca61ef4c5a831f43bdf9118252b58ad3ef92b4496537b5d389b2ecfc7a7d
+size 102612