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": 77051.74595250555,
-      "required_pages_per_second_upper_estimate": 0.006678650053298455
+      "required_token_positions_per_second": 6199.964581927836,
+      "required_pages_per_second_upper_estimate": 0.0005373972162949332
     },
     {
       "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": 9631.468244063193,
-      "required_pages_per_second_upper_estimate": 0.003842136685839793
+      "required_token_positions_per_second": 774.9955727409795,
+      "required_pages_per_second_upper_estimate": 0.00030915732118277467
     },
     {
       "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": 19262.936488126386,
-      "required_pages_per_second_upper_estimate": 0.004456141340424686
+      "required_token_positions_per_second": 1549.991145481959,
+      "required_pages_per_second_upper_estimate": 0.00035856317259478124
     },
     {
       "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": 28894.40473218958,
-      "required_pages_per_second_upper_estimate": 0.0017595176509647666
+      "required_token_positions_per_second": 2324.986718222938,
+      "required_pages_per_second_upper_estimate": 0.00014157949287719782
     },
     {
       "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": 28894.40473218958,
-      "required_pages_per_second_upper_estimate": 0.0022243712611155096
+      "required_token_positions_per_second": 2324.986718222938,
+      "required_pages_per_second_upper_estimate": 0.00017898391354395857
     },
     {
       "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": 28894.40473218958,
-      "required_pages_per_second_upper_estimate": 0.35877180342190024
+      "estimated_pages_for_buffer": 3,
+      "required_token_positions_per_second": 2324.986718222938,
+      "required_pages_per_second_upper_estimate": 0.028868553810334853
     }
   ],
   "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": 1.1856436634999739,
-    "producer_p95_seconds": 2.358020696999972,
-    "gpu_update_seconds": 0.9568634569999972,
-    "producer_batches_per_second": 0.8434237290553547,
-    "gpu_batches_per_second": 1.0450811896769958,
-    "producer_headroom": 0.8070413450997356,
-    "sustainable": false,
-    "buffer_seconds": 60.28239779099982,
-    "buffer_drain_seconds": 312.4109557156601,
+    "producer_batch_seconds": 0.04484002784127077,
+    "producer_p95_seconds": 0.11620487900000853,
+    "gpu_update_seconds": 11.891680835549998,
+    "producer_batches_per_second": 22.30150265606213,
+    "gpu_batches_per_second": 0.08409240155609586,
+    "producer_headroom": 265.2023517390614,
+    "sustainable": true,
+    "buffer_seconds": 71.35008501329999,
+    "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 7,
-    "max_pending_pages": 1726,
+    "upload_pages_per_commit": 21,
+    "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,
     "upload_sustainable": null,

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -0,0 +1 @@
+{"step": 315100, "loss": 0.5646674651652575, "lr": 0.0003, "grad_norm": 0.17352291347570195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46121906842, "tokens_per_second": 46362.37927092308, "data_wait_s": 7.901999993009667e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:46954b1df954862d20a91d0d94cf8e6d6c6424f64c1ec7dafe2dcaff34a00bef
-size 165
+oid sha256:a84f8b48832f1e6c69832efa8f3d6de12934304cb33bc2a9f28a19aa4c95b95a
+size 2283