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": 75458.65267281262,
-      "required_pages_per_second_upper_estimate": 0.006540564765472674
+      "required_token_positions_per_second": 75689.28670793981,
+      "required_pages_per_second_upper_estimate": 0.006560555539100882
     },
     {
       "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": 9432.331584101577,
-      "required_pages_per_second_upper_estimate": 0.003762698094822713
+      "required_token_positions_per_second": 9461.160838492477,
+      "required_pages_per_second_upper_estimate": 0.0037741985154350075
     },
     {
       "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": 18864.663168203155,
-      "required_pages_per_second_upper_estimate": 0.004364007817231477
+      "required_token_positions_per_second": 18922.321676984953,
+      "required_pages_per_second_upper_estimate": 0.00437734609848305
     },
     {
       "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": 28296.994752304734,
-      "required_pages_per_second_upper_estimate": 0.0017231385175577071
+      "required_token_positions_per_second": 28383.482515477433,
+      "required_pages_per_second_upper_estimate": 0.0017284051685687001
     },
     {
       "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": 28296.994752304734,
-      "required_pages_per_second_upper_estimate": 0.002178380987127306
+      "required_token_positions_per_second": 28383.482515477433,
+      "required_pages_per_second_upper_estimate": 0.002185039054549792
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28296.994752304734,
-      "required_pages_per_second_upper_estimate": 0.35135397087431536
+      "required_token_positions_per_second": 28383.482515477433,
+      "required_pages_per_second_upper_estimate": 0.35242785943699706
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.0511901649687303,
-    "producer_p95_seconds": 0.18669305799994618,
-    "gpu_update_seconds": 0.9770648877032472,
-    "producer_batches_per_second": 19.535002487506215,
-    "gpu_batches_per_second": 1.0234734791776885,
-    "producer_headroom": 19.086965011737917,
+    "producer_batch_seconds": 0.09841888091406048,
+    "producer_p95_seconds": 0.18062336000002688,
+    "gpu_update_seconds": 0.9740876576693372,
+    "producer_batches_per_second": 10.160652008156866,
+    "gpu_batches_per_second": 1.0266016534822566,
+    "producer_headroom": 9.897365715018768,
     "sustainable": true,
-    "buffer_seconds": 60.57802303760133,
+    "buffer_seconds": 60.39343477549891,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 395163.83161593793,
-    "upload_bytes_per_second": 5372382.916989888,
+    "generation_bytes_per_second": 393231.04518796876,
+    "upload_bytes_per_second": 5251888.749101955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.79295932158092,
+    "estimated_full_shards_per_hour": 84.37822834710404,
     "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": [
@@ -418,6 +418,18 @@
       "bytes": 11609605,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.5608764529997643,
+      "bytes": 2529116,
+      "pages": 35,
+      "shards": 2
+    },
+    {
+      "seconds": 1.0529914669996288,
+      "bytes": 5481895,
+      "pages": 19,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -48,3 +48,4 @@
 {"step": 319800, "loss": 0.6139744222164154, "lr": 0.0003, "grad_norm": 0.16847413332268588, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46809873976, "tokens_per_second": 139570.27631759262, "data_wait_s": 1.222300033987267e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 319900, "loss": 0.5427915640175343, "lr": 0.0003, "grad_norm": 0.1438469288188957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46824557290, "tokens_per_second": 148767.28337406035, "data_wait_s": 1.2318999324634206e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 320000, "loss": 0.5799524243921041, "lr": 0.0003, "grad_norm": 0.15063794863403282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46839154775, "tokens_per_second": 148510.84042039752, "data_wait_s": 1.4604999705625232e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 320100, "loss": 0.6229025758802891, "lr": 0.0003, "grad_norm": 0.1792688815400863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46853753312, "tokens_per_second": 131315.68225631822, "data_wait_s": 1.069899917638395e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9db643bb758edbe072ac7c5bc2503f5b8b13ea8181694beb8585fd9ac63ad2a0
-size 106848
+oid sha256:8cfc7cf7ca6464c69e779a7e5d2bf4c6d6eebd9a0f3f56a70ff411e87abea21b
+size 108966