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": 77025.68205668553,
-      "required_pages_per_second_upper_estimate": 0.005328815408887267
+      "required_token_positions_per_second": 76794.66926566833,
+      "required_pages_per_second_upper_estimate": 0.005312833407981172
     },
     {
       "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": 9628.210257085691,
-      "required_pages_per_second_upper_estimate": 0.004031130574748329
+      "required_token_positions_per_second": 9599.333658208541,
+      "required_pages_per_second_upper_estimate": 0.004019040545810421
     },
     {
       "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": 19256.420514171383,
-      "required_pages_per_second_upper_estimate": 0.004010604567143123
+      "required_token_positions_per_second": 19198.667316417082,
+      "required_pages_per_second_upper_estimate": 0.003998576099105149
     },
     {
       "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": 28884.630771257074,
-      "required_pages_per_second_upper_estimate": 0.0017315895104593943
+      "required_token_positions_per_second": 28798.000974625622,
+      "required_pages_per_second_upper_estimate": 0.0017263961864273792
     },
     {
       "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": 28884.630771257074,
-      "required_pages_per_second_upper_estimate": 0.0022456529924015393
+      "required_token_positions_per_second": 28798.000974625622,
+      "required_pages_per_second_upper_estimate": 0.0022389179067576494
     },
     {
       "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": 28884.630771257074,
-      "required_pages_per_second_upper_estimate": 0.3705010296334972
+      "required_token_positions_per_second": 28798.000974625622,
+      "required_pages_per_second_upper_estimate": 0.36938983561813754
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09037926416409192,
-    "producer_p95_seconds": 0.14240372499989462,
-    "gpu_update_seconds": 0.9571872397798611,
-    "producer_batches_per_second": 11.06448485998301,
-    "gpu_batches_per_second": 1.0447276754650274,
-    "producer_headroom": 10.5907837227132,
+    "producer_batch_seconds": 0.11747127801563906,
+    "producer_p95_seconds": 0.1434927450000032,
+    "gpu_update_seconds": 0.9600666388045855,
+    "producer_batches_per_second": 8.512719167547228,
+    "gpu_batches_per_second": 1.0415943639549199,
+    "producer_headroom": 8.172777678274437,
     "sustainable": true,
-    "buffer_seconds": 60.30279610613125,
+    "buffer_seconds": 60.48419824468889,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 262327.5927222717,
-    "upload_bytes_per_second": 5087751.603311389,
+    "generation_bytes_per_second": 268530.1835936344,
+    "upload_bytes_per_second": 4968346.39347336,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 56.28939472437966,
+    "estimated_full_shards_per_hour": 57.62032633644842,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 11801808,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0429953859998022,
+      "bytes": 2436103,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -21,3 +21,4 @@
 {"step": 606500, "loss": 0.5316051244735718, "lr": 0.0001, "grad_norm": 0.14572424153445881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88773619924, "tokens_per_second": 123176.79315265315, "data_wait_s": 8.829999842419056e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 606600, "loss": 0.5360864996910095, "lr": 0.0001, "grad_norm": 0.1538511914460197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88788296747, "tokens_per_second": 152232.94751619175, "data_wait_s": 1.1154999810969457e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 606700, "loss": 0.5418317914009094, "lr": 0.0001, "grad_norm": 0.16878725848738357, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88802910141, "tokens_per_second": 151640.92633213598, "data_wait_s": 1.0528000075282762e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 606800, "loss": 0.5774630308151245, "lr": 0.0001, "grad_norm": 0.1533205028361282, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88817671208, "tokens_per_second": 153215.73447032133, "data_wait_s": 1.502800023445161e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:581d6e82835111aca1756e68951904d65987affceb9fc834b1dc67f3c93dc07b
-size 49662
+oid sha256:f2f381a03c30a370f9a6af58082d97891cf7e519a1ae1b246f1fe879371978af
+size 51780