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": 76324.50726259945,
-      "required_pages_per_second_upper_estimate": 0.00528030650968272
+      "required_token_positions_per_second": 76440.09280479242,
+      "required_pages_per_second_upper_estimate": 0.005288302985687037
     },
     {
       "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": 9540.563407824931,
-      "required_pages_per_second_upper_estimate": 0.003994434669237188
+      "required_token_positions_per_second": 9555.011600599053,
+      "required_pages_per_second_upper_estimate": 0.004000483825839139
     },
     {
       "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": 19081.126815649863,
-      "required_pages_per_second_upper_estimate": 0.003974095512546791
+      "required_token_positions_per_second": 19110.023201198106,
+      "required_pages_per_second_upper_estimate": 0.003980113867607558
     },
     {
       "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": 28621.69022347479,
-      "required_pages_per_second_upper_estimate": 0.001715826626106816
+      "required_token_positions_per_second": 28665.03480179716,
+      "required_pages_per_second_upper_estimate": 0.0017184250673939032
     },
     {
       "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": 28621.69022347479,
-      "required_pages_per_second_upper_estimate": 0.0022252105213647084
+      "required_token_positions_per_second": 28665.03480179716,
+      "required_pages_per_second_upper_estimate": 0.0022285803716766213
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28621.69022347479,
-      "required_pages_per_second_upper_estimate": 0.3671283106101101
+      "required_token_positions_per_second": 28665.03480179716,
+      "required_pages_per_second_upper_estimate": 0.36768428832104716
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1041454026014037,
-    "producer_p95_seconds": 0.25335951799934264,
-    "gpu_update_seconds": 0.965980687518021,
-    "producer_batches_per_second": 9.601960096378962,
-    "gpu_batches_per_second": 1.0352173836615592,
-    "producer_headroom": 9.275308015420752,
+    "producer_batch_seconds": 0.09036237860161123,
+    "producer_p95_seconds": 0.257491027998185,
+    "gpu_update_seconds": 0.9645200220816269,
+    "producer_batches_per_second": 11.066552424530459,
+    "gpu_batches_per_second": 1.0367851129122236,
+    "producer_headroom": 10.6739113888756,
     "sustainable": true,
-    "buffer_seconds": 60.856783313635326,
+    "buffer_seconds": 60.76476139114249,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 277508.9653682234,
-    "upload_bytes_per_second": 5026151.619525781,
+    "generation_bytes_per_second": 278055.36931551894,
+    "upload_bytes_per_second": 5053937.820772797,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.54696388993288,
+    "estimated_full_shards_per_hour": 59.6642094573896,
     "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": 1.1245960120013478,
-      "bytes": 2666345,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2064912770001683,
       "bytes": 6004944,
@@ -562,6 +556,12 @@
       "bytes": 7075351,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.4313278709996666,
+      "bytes": 6586683,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -186,3 +186,4 @@
 {"step": 738600, "loss": 0.5704699158668518, "lr": 5e-05, "grad_norm": 0.26998518373392777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108110739457, "tokens_per_second": 130833.35308083435, "data_wait_s": 1.2314998457441106e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738700, "loss": 0.5004143714904785, "lr": 5e-05, "grad_norm": 0.22349828997716606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108125463221, "tokens_per_second": 151822.54949203963, "data_wait_s": 1.803800114430487e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 738800, "loss": 0.5550748109817505, "lr": 5e-05, "grad_norm": 0.2248784547664171, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108140265057, "tokens_per_second": 151710.9978331769, "data_wait_s": 1.7578000552020967e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 738900, "loss": 0.5104029178619385, "lr": 5e-05, "grad_norm": 0.22732493032322712, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108155032101, "tokens_per_second": 151513.5468122271, "data_wait_s": 1.7752001440385357e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e665de1f2bfaaa1dfab7d98ac4a70bbf92a7671adfbdc828ca435d40c1bf87bc
-size 400698
+oid sha256:732c8c8c1672ac8e500256cdce1a1a3d5fa7f4cd0ccc5e469f9816ebf99039cb
+size 402816