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": 75434.62101990751,
-      "required_pages_per_second_upper_estimate": 0.005218742114592731
+      "required_token_positions_per_second": 76647.33696221054,
+      "required_pages_per_second_upper_estimate": 0.005302640617369919
     },
     {
       "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": 9429.32762748844,
-      "required_pages_per_second_upper_estimate": 0.003947862570877534
+      "required_token_positions_per_second": 9580.917120276317,
+      "required_pages_per_second_upper_estimate": 0.0040113299259592426
     },
     {
       "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": 18858.65525497688,
-      "required_pages_per_second_upper_estimate": 0.003927760553428201
+      "required_token_positions_per_second": 19161.834240552635,
+      "required_pages_per_second_upper_estimate": 0.003990904740756115
     },
     {
       "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": 28287.98288246532,
-      "required_pages_per_second_upper_estimate": 0.0016958213805549046
+      "required_token_positions_per_second": 28742.751360828956,
+      "required_pages_per_second_upper_estimate": 0.0017230840564415447
     },
     {
       "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": 28287.98288246532,
-      "required_pages_per_second_upper_estimate": 0.0021992662434246906
+      "required_token_positions_per_second": 28742.751360828956,
+      "required_pages_per_second_upper_estimate": 0.002234622492302327
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28287.98288246532,
-      "required_pages_per_second_upper_estimate": 0.36284787114666717
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28742.751360828956,
+      "required_pages_per_second_upper_estimate": 0.36868115289476733
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09448133852342266,
-    "producer_p95_seconds": 0.26248719899740536,
-    "gpu_update_seconds": 0.977376157037269,
-    "producer_batches_per_second": 10.584100687270558,
-    "gpu_batches_per_second": 1.0231475290243532,
-    "producer_headroom": 10.344647655420015,
+    "producer_batch_seconds": 0.0712423114766807,
+    "producer_p95_seconds": 0.25802149399896734,
+    "gpu_update_seconds": 0.9619120888224745,
+    "producer_batches_per_second": 14.036602396418928,
+    "gpu_batches_per_second": 1.039596041696649,
+    "producer_headroom": 13.501977531109882,
     "sustainable": true,
-    "buffer_seconds": 60.597321736310676,
+    "buffer_seconds": 60.600461595815894,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 300622.1024057796,
-    "upload_bytes_per_second": 3880872.8627244416,
+    "generation_bytes_per_second": 358935.0455837974,
+    "upload_bytes_per_second": 3809073.9940243065,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.50650505189934,
+    "estimated_full_shards_per_hour": 77.01910520205918,
     "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": 1.1427723569995578,
-      "bytes": 13775084,
-      "pages": 72,
-      "shards": 3
-    },
     {
       "seconds": 1.2821505469983094,
       "bytes": 1156096,
@@ -562,6 +556,12 @@
       "bytes": 2631414,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.286528674998408,
+      "bytes": 6199355,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -216,3 +216,4 @@
 {"step": 666600, "loss": 0.5367074012756348, "lr": 0.0001, "grad_norm": 0.19993975317060272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97573544513, "tokens_per_second": 151572.8148915651, "data_wait_s": 1.4500998076982796e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 666700, "loss": 0.5609776973724365, "lr": 0.0001, "grad_norm": 0.1883014932611547, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97588053446, "tokens_per_second": 150949.39067962268, "data_wait_s": 8.032002369873226e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 666800, "loss": 0.5390537977218628, "lr": 0.0001, "grad_norm": 0.1958525864032979, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97602555346, "tokens_per_second": 151428.08310205027, "data_wait_s": 9.765000868355855e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 666900, "loss": 0.5449493527412415, "lr": 0.0001, "grad_norm": 0.18722814729427356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 97617364320, "tokens_per_second": 155077.09781617267, "data_wait_s": 1.1454998457338661e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47f335a8635d9313b9ff613fadc8456797e5300d4e741a6ba25f0eb74143bd50
-size 465021
+oid sha256:62ed930afd86264f64dc3c84af9a3451dc08eefddf4bbc0e4703ba1e4ffd1782
+size 467139