Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77144.5045607611,
-      "required_pages_per_second_upper_estimate": 0.005337035825425409
+      "required_token_positions_per_second": 75995.66133108345,
+      "required_pages_per_second_upper_estimate": 0.005257556185112757
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9643.063070095137,
-      "required_pages_per_second_upper_estimate": 0.004037349137393378
+      "required_token_positions_per_second": 9499.457666385431,
+      "required_pages_per_second_upper_estimate": 0.003977224553681961
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19286.126140190274,
-      "required_pages_per_second_upper_estimate": 0.00401679146565282
+      "required_token_positions_per_second": 18998.915332770863,
+      "required_pages_per_second_upper_estimate": 0.003956973028725695
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28929.18921028541,
-      "required_pages_per_second_upper_estimate": 0.0017342607208423462
+      "required_token_positions_per_second": 28498.372999156298,
+      "required_pages_per_second_upper_estimate": 0.0017084339468034213
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28929.18921028541,
-      "required_pages_per_second_upper_estimate": 0.0022491172150441324
+      "required_token_positions_per_second": 28498.372999156298,
+      "required_pages_per_second_upper_estimate": 0.002215623149589091
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28929.18921028541,
-      "required_pages_per_second_upper_estimate": 0.3710725774462284
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28498.372999156298,
+      "required_pages_per_second_upper_estimate": 0.36554652966427187
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09812232820308964,
-    "producer_p95_seconds": 0.34375213400198845,
-    "gpu_update_seconds": 0.9557129236850544,
-    "producer_batches_per_second": 10.191360298037775,
-    "gpu_batches_per_second": 1.0463393088210868,
-    "producer_headroom": 9.74001474676547,
+    "producer_batch_seconds": 0.6251384444220776,
+    "producer_p95_seconds": 0.35355271399748744,
+    "gpu_update_seconds": 0.9701606474453306,
+    "producer_batches_per_second": 1.5996456607695455,
+    "gpu_batches_per_second": 1.030757125258836,
+    "producer_headroom": 1.5519132699352958,
     "sustainable": true,
-    "buffer_seconds": 60.20991419215843,
+    "buffer_seconds": 60.1499601416105,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 406618.36189258524,
-    "upload_bytes_per_second": 5126782.612468387,
+    "generation_bytes_per_second": 358794.4000004342,
+    "upload_bytes_per_second": 4903945.440431726,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.25083487113159,
+    "estimated_full_shards_per_hour": 76.98892593393106,
     "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": [
-    {
-      "seconds": 1.0651401020004414,
-      "bytes": 11486452,
-      "pages": 47,
-      "shards": 3
-    },
     {
       "seconds": 1.0763056159994449,
       "bytes": 2441249,
@@ -562,6 +556,12 @@
       "bytes": 7438556,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 2.768625452001288,
+      "bytes": 924425,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -388,3 +388,4 @@
 {"step": 435900, "loss": 0.5809106826782227, "lr": 0.0003, "grad_norm": 0.2027705148394604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63803264435, "tokens_per_second": 151935.50790462716, "data_wait_s": 1.3031996786594391e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436000, "loss": 0.4993954300880432, "lr": 0.0003, "grad_norm": 0.14581099757722213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63817751974, "tokens_per_second": 150349.26657984426, "data_wait_s": 1.1990996426902711e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 436100, "loss": 0.5766322016716003, "lr": 0.0003, "grad_norm": 0.15279914796720345, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63832285536, "tokens_per_second": 151045.89412842927, "data_wait_s": 2.7783004043158144e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 436200, "loss": 0.5508224964141846, "lr": 0.0003, "grad_norm": 0.15514427238020523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63846978055, "tokens_per_second": 136587.17837558212, "data_wait_s": 1.2106000212952495e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9457fe292a7a998f1c2a67d262424e52617ef865c186144d79b7a64cfd603d2
-size 832449
+oid sha256:500c32a6f8ce3e9c0b1ace720f87f785ad21cecc738f01c1cc39c3bd49c062b6
+size 834567