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": 74505.36266274125,
-      "required_pages_per_second_upper_estimate": 0.006457936003491129
+      "required_token_positions_per_second": 75922.93218729971,
+      "required_pages_per_second_upper_estimate": 0.006580807337082744
     },
     {
       "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": 9313.170332842656,
-      "required_pages_per_second_upper_estimate": 0.0037151628900760555
+      "required_token_positions_per_second": 9490.366523412464,
+      "required_pages_per_second_upper_estimate": 0.003785849099813493
     },
     {
       "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": 18626.34066568531,
-      "required_pages_per_second_upper_estimate": 0.004308876100606765
+      "required_token_positions_per_second": 18980.733046824927,
+      "required_pages_per_second_upper_estimate": 0.0043908585408905295
     },
     {
       "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": 27939.510998527967,
-      "required_pages_per_second_upper_estimate": 0.0017013696325250062
+      "required_token_positions_per_second": 28471.099570237388,
+      "required_pages_per_second_upper_estimate": 0.0017337405875124218
     },
     {
       "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": 27939.510998527967,
-      "required_pages_per_second_upper_estimate": 0.002150860898183204
+      "required_token_positions_per_second": 28471.099570237388,
+      "required_pages_per_second_upper_estimate": 0.002191784057964744
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27939.510998527967,
-      "required_pages_per_second_upper_estimate": 0.34691521907356826
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28471.099570237388,
+      "required_pages_per_second_upper_estimate": 0.3535157700216967
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.3090939151872476,
-    "producer_p95_seconds": 0.2605821119977918,
-    "gpu_update_seconds": 0.9895663528777106,
-    "producer_batches_per_second": 3.235262652757835,
-    "gpu_batches_per_second": 1.0105436559074061,
-    "producer_headroom": 3.2015070638910377,
+    "producer_batch_seconds": 0.11411454092188933,
+    "producer_p95_seconds": 0.2691598500023247,
+    "gpu_update_seconds": 0.9710899971317642,
+    "producer_batches_per_second": 8.76312511903714,
+    "gpu_batches_per_second": 1.0297706731133314,
+    "producer_headroom": 8.509783146711067,
     "sustainable": true,
-    "buffer_seconds": 60.36354752554034,
+    "buffer_seconds": 60.207579822169386,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 424995.6776533209,
-    "upload_bytes_per_second": 5755775.831788403,
+    "generation_bytes_per_second": 441606.12367167394,
+    "upload_bytes_per_second": 5831001.606646914,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.1941790313694,
+    "estimated_full_shards_per_hour": 94.75839407551445,
     "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": 1.196992418001173,
-      "bytes": 774765,
-      "pages": 25,
-      "shards": 1
-    },
     {
       "seconds": 1.1212218569999095,
       "bytes": 7621728,
@@ -562,6 +556,12 @@
       "bytes": 10294757,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.5211265639991325,
+      "bytes": 8178789,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -187,3 +187,4 @@
 {"step": 373700, "loss": 0.5965260341763496, "lr": 0.0003, "grad_norm": 0.1407083360208807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54698785941, "tokens_per_second": 150111.3880085052, "data_wait_s": 1.3315999240148813e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 373800, "loss": 0.5403846129775047, "lr": 0.0003, "grad_norm": 0.2319163103574092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54713423612, "tokens_per_second": 148937.27212668382, "data_wait_s": 9.621999197406694e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 373900, "loss": 0.5923236533999443, "lr": 0.0003, "grad_norm": 0.16723194428624796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54727941165, "tokens_per_second": 148546.19985122856, "data_wait_s": 1.1596999684115872e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 374000, "loss": 0.5675775744020939, "lr": 0.0003, "grad_norm": 0.16214766662555807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54742535343, "tokens_per_second": 149163.61020735494, "data_wait_s": 1.4774999726796523e-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:a590c25ae883621b72b07ed4d801bc146b767ce3757d33a802940fa656cbdf78
-size 402816
+oid sha256:5c9ba50e314c51d651081222a3d5cf69a3ef2a59c0db36fa28825c1728a114ba
+size 404934