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": 6909.4253631529855,
-      "required_pages_per_second_upper_estimate": 0.0005988914786996183
+      "required_token_positions_per_second": 72112.39904821498,
+      "required_pages_per_second_upper_estimate": 0.006250519982294826
     },
     {
       "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": 863.6781703941232,
-      "required_pages_per_second_upper_estimate": 0.0003445341353096071
+      "required_token_positions_per_second": 9014.049881026873,
+      "required_pages_per_second_upper_estimate": 0.003595839269597444
     },
     {
       "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": 1727.3563407882464,
-      "required_pages_per_second_upper_estimate": 0.00039959348900806663
+      "required_token_positions_per_second": 18028.099762053746,
+      "required_pages_per_second_upper_estimate": 0.004170483596231907
     },
     {
       "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": 2591.0345111823694,
-      "required_pages_per_second_upper_estimate": 0.0001577804076235341
+      "required_token_positions_per_second": 27042.149643080615,
+      "required_pages_per_second_upper_estimate": 0.001646725034069434
     },
     {
       "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": 2591.0345111823694,
-      "required_pages_per_second_upper_estimate": 0.0001994650091134025
+      "required_token_positions_per_second": 27042.149643080615,
+      "required_pages_per_second_upper_estimate": 0.0020817795369856484
     },
     {
       "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": 2591.0345111823694,
-      "required_pages_per_second_upper_estimate": 0.03217197699420601
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27042.149643080615,
+      "required_pages_per_second_upper_estimate": 0.3357729943141738
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,20 +155,20 @@
   "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": 717798,
-    "producer_batch_seconds": 0.06805186365079317,
+    "producer_batch_seconds": 0.06380424955660066,
     "producer_p95_seconds": 0.15090544800000316,
-    "gpu_update_seconds": 10.670641352200036,
-    "producer_batches_per_second": 14.694674713560833,
-    "gpu_batches_per_second": 0.09371507925283455,
-    "producer_headroom": 156.80160365565044,
+    "gpu_update_seconds": 1.022403927384316,
+    "producer_batches_per_second": 15.672937256520843,
+    "gpu_batches_per_second": 0.9780870096600338,
+    "producer_headroom": 16.024072604714878,
     "sustainable": true,
-    "buffer_seconds": 64.02384811320022,
+    "buffer_seconds": 60.32183171567464,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 17,
+    "upload_pages_per_commit": 21,
     "max_pending_pages": 4096,
     "generation_bytes_per_second": 0,
     "upload_bytes_per_second": null,

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -1 +1,2 @@
 {"step": 355100, "loss": 0.6090846359729767, "lr": 0.0003, "grad_norm": 0.15904267394337226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51976574318, "tokens_per_second": 50459.87599969507, "data_wait_s": 9.298000009039242e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 355200, "loss": 0.5521676354110241, "lr": 0.0003, "grad_norm": 0.14981194429073466, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51991218815, "tokens_per_second": 150210.32586298898, "data_wait_s": 1.2212999990879325e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6175406ac3f10f930f2763ba6028f056689fe013f1095bb050be6c49da66fe4
-size 2283
+oid sha256:5fd2dc461dceef696334d28d89f044d54d1db07fbbc8d364c48bf3686cb081c7
+size 4401