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": 77183.6572876782,
-      "required_pages_per_second_upper_estimate": 0.006690083793504997
+      "required_token_positions_per_second": 6402.400470292964,
+      "required_pages_per_second_upper_estimate": 0.0005549438460293542
     },
     {
       "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": 9647.957160959775,
-      "required_pages_per_second_upper_estimate": 0.0038487143613211167
+      "required_token_positions_per_second": 800.3000587866205,
+      "required_pages_per_second_upper_estimate": 0.00031925165900216235
     },
     {
       "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": 19295.91432191955,
-      "required_pages_per_second_upper_estimate": 0.004463770181882682
+      "required_token_positions_per_second": 1600.600117573241,
+      "required_pages_per_second_upper_estimate": 0.0003702706676006113
     },
     {
       "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": 28943.871482879324,
-      "required_pages_per_second_upper_estimate": 0.0017625299165497888
+      "required_token_positions_per_second": 2400.9001763598617,
+      "required_pages_per_second_upper_estimate": 0.00014620222419060288
     },
     {
       "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": 28943.871482879324,
-      "required_pages_per_second_upper_estimate": 0.002228179348516336
+      "required_token_positions_per_second": 2400.9001763598617,
+      "required_pages_per_second_upper_estimate": 0.00018482794169324073
     },
     {
       "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": 28943.871482879324,
-      "required_pages_per_second_upper_estimate": 0.3593860149109021
+      "estimated_pages_for_buffer": 3,
+      "required_token_positions_per_second": 2400.9001763598617,
+      "required_pages_per_second_upper_estimate": 0.029811144894394646
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,21 +155,21 @@
   "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": 63,
+    "prefetch_batches": 6,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 1.5474661249999713,
-    "producer_p95_seconds": 2.844574209999962,
-    "gpu_update_seconds": 0.9552281219999941,
-    "producer_batches_per_second": 0.6462176999189683,
-    "gpu_batches_per_second": 1.0468703516666422,
-    "producer_headroom": 0.6172853198967517,
-    "sustainable": false,
-    "buffer_seconds": 60.17937168599963,
-    "buffer_drain_seconds": 157.24343699009538,
+    "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,
+    "sustainable": true,
+    "buffer_seconds": 69.09408464099995,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 7,
-    "max_pending_pages": 1768,
+    "upload_pages_per_commit": 17,
+    "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,
     "upload_sustainable": null,

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

@@ -0,0 +1 @@
+{"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}

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 sha256:38c569c5bbef71d5c06408c2589ac140325223cbe25d24d26270c3c2aec9f6be
-size 165
+oid sha256:170ce148228d7486d6f03fb302b445ea9254996c56b461a29d2b64251ac88226
+size 2283