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": 74853.16600443146,
-      "required_pages_per_second_upper_estimate": 0.006488082715649327
+      "required_token_positions_per_second": 75351.08193525887,
+      "required_pages_per_second_upper_estimate": 0.0065312408065770575
     },
     {
       "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": 9356.645750553933,
-      "required_pages_per_second_upper_estimate": 0.0037325058842165044
+      "required_token_positions_per_second": 9418.88524190736,
+      "required_pages_per_second_upper_estimate": 0.0037573341478807083
     },
     {
       "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": 18713.291501107866,
-      "required_pages_per_second_upper_estimate": 0.004328990646099334
+      "required_token_positions_per_second": 18837.77048381472,
+      "required_pages_per_second_upper_estimate": 0.004357786667993292
     },
     {
       "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": 28069.9372516618,
-      "required_pages_per_second_upper_estimate": 0.0017093119070471913
+      "required_token_positions_per_second": 28256.655725722077,
+      "required_pages_per_second_upper_estimate": 0.0017206820824813393
     },
     {
       "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": 28069.9372516618,
-      "required_pages_per_second_upper_estimate": 0.002160901472192423
+      "required_token_positions_per_second": 28256.655725722077,
+      "required_pages_per_second_upper_estimate": 0.0021752755771954004
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28069.9372516618,
-      "required_pages_per_second_upper_estimate": 0.348534676628901
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28256.655725722077,
+      "required_pages_per_second_upper_estimate": 0.350853095170196
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.9141480985548469,
-    "producer_p95_seconds": 0.3989767169987317,
-    "gpu_update_seconds": 0.9849683578599087,
-    "producer_batches_per_second": 1.0939146529767705,
-    "gpu_batches_per_second": 1.0152610406417029,
-    "producer_headroom": 1.0774713193814216,
+    "producer_batch_seconds": 0.12465785078950375,
+    "producer_p95_seconds": 0.3989656740013743,
+    "gpu_update_seconds": 0.9784597394812006,
+    "producer_batches_per_second": 8.02195765181763,
+    "gpu_batches_per_second": 1.0220144576722396,
+    "producer_headroom": 7.849162594126701,
     "sustainable": true,
-    "buffer_seconds": 60.08306982945443,
+    "buffer_seconds": 60.66450384783444,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 251892.62602489186,
-    "upload_bytes_per_second": 5442384.376229002,
+    "generation_bytes_per_second": 251758.896143092,
+    "upload_bytes_per_second": 5622072.318178858,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 54.050293784714384,
+    "estimated_full_shards_per_hour": 54.02159846515245,
     "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": 1.6717624239972793,
-      "bytes": 5422811,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.1521110989997396,
       "bytes": 12146138,
@@ -562,6 +556,12 @@
       "bytes": 678366,
       "pages": 20,
       "shards": 1
+    },
+    {
+      "seconds": 1.3286253929982195,
+      "bytes": 17203756,
+      "pages": 37,
+      "shards": 5
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -358,3 +358,4 @@
 {"step": 150800, "loss": 0.5710970014333725, "lr": 0.0003, "grad_norm": 0.14444581917404686, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22069015443, "tokens_per_second": 150694.64432425157, "data_wait_s": 1.0910996934399009e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 150900, "loss": 0.5997108649462461, "lr": 0.0003, "grad_norm": 0.139786398744748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22083767976, "tokens_per_second": 105962.76062197078, "data_wait_s": 1.3844000932294875e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 151000, "loss": 0.6267396453768015, "lr": 0.0003, "grad_norm": 0.1518755776149157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22098506659, "tokens_per_second": 149930.15825412446, "data_wait_s": 1.4491000911220908e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 151100, "loss": 0.5722631234675646, "lr": 0.0003, "grad_norm": 0.14709183133475207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22113193280, "tokens_per_second": 149919.23307072968, "data_wait_s": 1.3406999642029405e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5148efe248acda601511e0d03fc87228b14f0711ae955f9b0843b7d72ed57033
-size 768126
+oid sha256:110f2b32e4c1a7589979079d7209cf7fb0a19a789483f14836b815a1476c041a
+size 770244