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": 75674.59292809678,
-      "required_pages_per_second_upper_estimate": 0.006559281919505133
+      "required_token_positions_per_second": 74475.6435401087,
+      "required_pages_per_second_upper_estimate": 0.006455360025263768
     },
     {
       "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": 9459.324116012098,
-      "required_pages_per_second_upper_estimate": 0.003773465819376136
+      "required_token_positions_per_second": 9309.455442513587,
+      "required_pages_per_second_upper_estimate": 0.0037136809647812298
     },
     {
       "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": 18918.648232024196,
-      "required_pages_per_second_upper_estimate": 0.004376496311641802
+      "required_token_positions_per_second": 18618.910885027173,
+      "required_pages_per_second_upper_estimate": 0.0043071573516111775
     },
     {
       "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": 28377.97234803629,
-      "required_pages_per_second_upper_estimate": 0.001728069628281149
+      "required_token_positions_per_second": 27928.36632754076,
+      "required_pages_per_second_upper_estimate": 0.0017006909805334551
     },
     {
       "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": 28377.97234803629,
-      "required_pages_per_second_upper_estimate": 0.002184614866607053
+      "required_token_positions_per_second": 27928.36632754076,
+      "required_pages_per_second_upper_estimate": 0.002150002950560185
     },
     {
       "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": 28377.97234803629,
-      "required_pages_per_second_upper_estimate": 0.35235944159872223
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27928.36632754076,
+      "required_pages_per_second_upper_estimate": 0.3467768395587216
     }
   ],
   "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": 0.1494978958984401,
-    "producer_p95_seconds": 0.12998108199997205,
-    "gpu_update_seconds": 0.9742767968379247,
-    "producer_batches_per_second": 6.689057354220825,
-    "gpu_batches_per_second": 1.0264023563381182,
-    "producer_headroom": 6.516993372935429,
+    "producer_batch_seconds": 0.04182273801562175,
+    "producer_p95_seconds": 0.1311279139999897,
+    "gpu_update_seconds": 0.9899612342428964,
+    "producer_batches_per_second": 23.91043837508862,
+    "gpu_batches_per_second": 1.0101405645088528,
+    "producer_headroom": 23.670407085091444,
     "sustainable": true,
-    "buffer_seconds": 60.40516140395133,
+    "buffer_seconds": 60.38763528881668,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 107,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 683486.9912240453,
-    "upload_bytes_per_second": 626714.046389553,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 639939.5303847975,
+    "upload_bytes_per_second": 2579234.318862371,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.33019877691754,
+    "estimated_full_shards_per_hour": 68.65806309536907,
     "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": [
@@ -202,6 +202,12 @@
       "bytes": 2340308,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.0020929300000034,
+      "bytes": 10862030,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -1,3 +1,4 @@
 {"step": 82000, "loss": 0.6297517661005259, "lr": 0.0003, "grad_norm": 0.13215398907681755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 11999942244, "tokens_per_second": 40155.054755347504, "data_wait_s": 1.227600000675011e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 82100, "loss": 0.6415744367986917, "lr": 0.0003, "grad_norm": 0.14695724674884675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12014676584, "tokens_per_second": 149211.18322307098, "data_wait_s": 1.2620999996215687e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 82200, "loss": 0.5415969882160425, "lr": 0.0003, "grad_norm": 0.13570390417133701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12029317961, "tokens_per_second": 149121.2710085824, "data_wait_s": 1.574399993842235e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 82300, "loss": 0.6576139517128468, "lr": 0.0003, "grad_norm": 0.12632208377102014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12044048685, "tokens_per_second": 150014.86912897477, "data_wait_s": 9.185999942928902e-06, "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:d393f16e62a0d28c3c31746cdf2e854ed3030fa909692e1487f0979c0bdd5d48
-size 6519
+oid sha256:d82362b8ab40d4b38e76593129caddaef2089103d65c73fdba11b538491ebb76
+size 8637