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": 75351.3341061124,
-      "required_pages_per_second_upper_estimate": 0.006531262664107519
+      "required_token_positions_per_second": 74874.27488094913,
+      "required_pages_per_second_upper_estimate": 0.00648991237956592
     },
     {
       "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": 9418.91676326405,
-      "required_pages_per_second_upper_estimate": 0.003757346722221178
+      "required_token_positions_per_second": 9359.284360118641,
+      "required_pages_per_second_upper_estimate": 0.003733558465022595
     },
     {
       "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": 18837.8335265281,
-      "required_pages_per_second_upper_estimate": 0.0043578012518155195
+      "required_token_positions_per_second": 18718.568720237283,
+      "required_pages_per_second_upper_estimate": 0.004330211437869041
     },
     {
       "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": 28256.75028979215,
-      "required_pages_per_second_upper_estimate": 0.0017206878409370674
+      "required_token_positions_per_second": 28077.853080355926,
+      "required_pages_per_second_upper_estimate": 0.0017097939394835203
     },
     {
       "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": 28256.75028979215,
-      "required_pages_per_second_upper_estimate": 0.002175282856999285
+      "required_token_positions_per_second": 28077.853080355926,
+      "required_pages_per_second_upper_estimate": 0.002161510854597713
     },
     {
       "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": 28256.75028979215,
-      "required_pages_per_second_upper_estimate": 0.35085426933946073
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28077.853080355926,
+      "required_pages_per_second_upper_estimate": 0.34863296472870764
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.15831881689831562,
-    "producer_p95_seconds": 0.19946176200028276,
-    "gpu_update_seconds": 0.9784564649668132,
-    "producer_batches_per_second": 6.31636857570933,
-    "gpu_batches_per_second": 1.0220178779583386,
-    "producer_headroom": 6.180291668016015,
+    "producer_batch_seconds": 0.08001460710160302,
+    "producer_p95_seconds": 0.204020387998753,
+    "gpu_update_seconds": 0.9846906713584642,
+    "producer_batches_per_second": 12.497718057031689,
+    "gpu_batches_per_second": 1.0155473481031512,
+    "producer_headroom": 12.306386384027334,
     "sustainable": true,
-    "buffer_seconds": 60.66430082794241,
+    "buffer_seconds": 60.06613095286632,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382341.62832227256,
-    "upload_bytes_per_second": 3674976.047394571,
+    "generation_bytes_per_second": 425305.97743218497,
+    "upload_bytes_per_second": 3628507.6118907486,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.04161298037656,
+    "estimated_full_shards_per_hour": 91.26076214050447,
     "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": 1.008947303000241,
-      "bytes": 7015503,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.0751419640000677,
       "bytes": 5874473,
@@ -562,6 +556,12 @@
       "bytes": 11647558,
       "pages": 33,
       "shards": 4
+    },
+    {
+      "seconds": 0.8895111490001,
+      "bytes": 1197830,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -107,3 +107,4 @@
 {"step": 247800, "loss": 0.5763643234968185, "lr": 0.0003, "grad_norm": 0.15381343837353498, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36271597498, "tokens_per_second": 148702.6365903454, "data_wait_s": 1.8566999642644078e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 247900, "loss": 0.6008405964821577, "lr": 0.0003, "grad_norm": 0.17136289145099753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36286179430, "tokens_per_second": 147734.6405579104, "data_wait_s": 1.2148000678280368e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 248000, "loss": 0.5956814158707857, "lr": 0.0003, "grad_norm": 0.15833271851660244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36300645496, "tokens_per_second": 147757.1834952601, "data_wait_s": 8.596000043326057e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 248100, "loss": 0.5804008096456528, "lr": 0.0003, "grad_norm": 0.14121031873210405, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36315310179, "tokens_per_second": 149589.7096732957, "data_wait_s": 1.3102000593789853e-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 sha256:29a4e0d542aed7efdfbf3138b6da2d68dc0a5f6f1f3fcb236648db2038993923
-size 232593
+oid sha256:864529868ce50eda4718bde7955dc351d550a3120156c1c736883897d2e019f1
+size 234711