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": 77310.54947496414,
-      "required_pages_per_second_upper_estimate": 0.005348523197867257
+      "required_token_positions_per_second": 76067.88245488766,
+      "required_pages_per_second_upper_estimate": 0.005262552610033626
     },
     {
       "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": 9663.818684370517,
-      "required_pages_per_second_upper_estimate": 0.004046039079663967
+      "required_token_positions_per_second": 9508.485306860957,
+      "required_pages_per_second_upper_estimate": 0.003981004238230493
     },
     {
       "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": 19327.637368741034,
-      "required_pages_per_second_upper_estimate": 0.004025437159835229
+      "required_token_positions_per_second": 19016.970613721915,
+      "required_pages_per_second_upper_estimate": 0.00396073346759802
     },
     {
       "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": 28991.45605311155,
-      "required_pages_per_second_upper_estimate": 0.0017379935229938092
+      "required_token_positions_per_second": 28525.455920582874,
+      "required_pages_per_second_upper_estimate": 0.001710057526589718
     },
     {
       "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": 28991.45605311155,
-      "required_pages_per_second_upper_estimate": 0.002253958188190968
+      "required_token_positions_per_second": 28525.455920582874,
+      "required_pages_per_second_upper_estimate": 0.0022177287276048253
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28991.45605311155,
-      "required_pages_per_second_upper_estimate": 0.3718712696490752
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28525.455920582874,
+      "required_pages_per_second_upper_estimate": 0.3658939203009566
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.1958027745703177,
-    "producer_p95_seconds": 0.1372530209996512,
-    "gpu_update_seconds": 0.9536602766466653,
-    "producer_batches_per_second": 5.10717992732465,
-    "gpu_batches_per_second": 1.0485914371061757,
-    "producer_headroom": 4.870514622376722,
+    "producer_batch_seconds": 0.20682684040624366,
+    "producer_p95_seconds": 0.13941991799993048,
+    "gpu_update_seconds": 0.9692395478962447,
+    "producer_batches_per_second": 4.834962416076304,
+    "gpu_batches_per_second": 1.0317366869423783,
+    "producer_headroom": 4.686236786253132,
     "sustainable": true,
-    "buffer_seconds": 60.080597428739914,
+    "buffer_seconds": 60.09285196956717,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 72,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 281412.9115702397,
-    "upload_bytes_per_second": 2760935.6732347663,
+    "generation_bytes_per_second": 397356.961545926,
+    "upload_bytes_per_second": 5354023.19577642,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.384659865669185,
+    "estimated_full_shards_per_hour": 85.2635539511045,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 6273746,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.1453777450001326,
+      "bytes": 12048299,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 526700, "loss": 0.60279381275177, "lr": 0.0003, "grad_norm": 0.201481649817476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77091369057, "tokens_per_second": 153328.03270316366, "data_wait_s": 1.0875000043597538e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526800, "loss": 0.6574923992156982, "lr": 0.0003, "grad_norm": 0.17117518403959395, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77106153896, "tokens_per_second": 153519.46429840563, "data_wait_s": 1.0234000001219101e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526900, "loss": 0.5783701539039612, "lr": 0.0003, "grad_norm": 0.14428829229489473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77120852088, "tokens_per_second": 152155.4686408566, "data_wait_s": 1.2045999937981833e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 527000, "loss": 0.6127535104751587, "lr": 0.0003, "grad_norm": 0.17756967529033216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77135404411, "tokens_per_second": 151332.22763061014, "data_wait_s": 1.0801000371429836e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:13fc062cb734aed88f95fc6fe175e56b041a23e3e8d6ef6db29b667603b3d243
-size 40407
+oid sha256:1e167c288a8731b76041801f0cbcc8a398aeec7300dd1fa68cbe9ae52a2829aa
+size 42525