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": 6402.400470292964,
-      "required_pages_per_second_upper_estimate": 0.0005549438460293542
+      "required_token_positions_per_second": 71775.01280807515,
+      "required_pages_per_second_upper_estimate": 0.006221276198097113
     },
     {
       "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": 800.3000587866205,
-      "required_pages_per_second_upper_estimate": 0.00031925165900216235
+      "required_token_positions_per_second": 8971.876601009393,
+      "required_pages_per_second_upper_estimate": 0.0035790157176517445
     },
     {
       "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": 1600.600117573241,
-      "required_pages_per_second_upper_estimate": 0.0003702706676006113
+      "required_token_positions_per_second": 17943.753202018786,
+      "required_pages_per_second_upper_estimate": 0.004150971504016575
     },
     {
       "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": 2400.9001763598617,
-      "required_pages_per_second_upper_estimate": 0.00014620222419060288
+      "required_token_positions_per_second": 26915.62980302818,
+      "required_pages_per_second_upper_estimate": 0.0016390206396085405
     },
     {
       "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": 2400.9001763598617,
-      "required_pages_per_second_upper_estimate": 0.00018482794169324073
+      "required_token_positions_per_second": 26915.62980302818,
+      "required_pages_per_second_upper_estimate": 0.002072039689469079
     },
     {
       "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": 3,
-      "required_token_positions_per_second": 2400.9001763598617,
-      "required_pages_per_second_upper_estimate": 0.029811144894394646
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 26915.62980302818,
+      "required_pages_per_second_upper_estimate": 0.33420204133538844
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "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": 6,
+    "prefetch_batches": 59,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06115566107936347,
-    "producer_p95_seconds": 0.3069990370000255,
-    "gpu_update_seconds": 11.515680773499993,
-    "producer_batches_per_second": 16.351715971188195,
-    "gpu_batches_per_second": 0.08683811401764546,
-    "producer_headroom": 188.30114122314467,
+    "producer_batch_seconds": 0.0523112050285693,
+    "producer_p95_seconds": 0.11531334599999354,
+    "gpu_update_seconds": 1.0272098480448495,
+    "producer_batches_per_second": 19.11636330025009,
+    "gpu_batches_per_second": 0.9735109159081373,
+    "producer_headroom": 19.63651664082003,
     "sustainable": true,
-    "buffer_seconds": 69.09408464099995,
+    "buffer_seconds": 60.60538103464612,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 17,
+    "upload_pages_per_commit": 24,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 379908.39896024216,
+    "upload_bytes_per_second": 54725.7885397608,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 81.51949860196541,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 0.8557939730000044,
+      "bytes": 46834,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

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

@@ -1 +1,2 @@
 {"step": 237200, "loss": 0.6311556715518236, "lr": 0.0003, "grad_norm": 0.13991984047282274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34718816495, "tokens_per_second": 21867.513315983448, "data_wait_s": 1.910299999963172e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 237300, "loss": 0.5835124477744102, "lr": 0.0003, "grad_norm": 0.15055686807409902, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34733571560, "tokens_per_second": 149880.6769248668, "data_wait_s": 8.639999919068941e-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 sha256:170ce148228d7486d6f03fb302b445ea9254996c56b461a29d2b64251ac88226
-size 2283
+oid sha256:f99df8f5f789e6ab5e535ce025e05986441266eeb78e71ac00d0ba848f4c53ab
+size 4401