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": 76614.16123552641,
-      "required_pages_per_second_upper_estimate": 0.00530034544361962
+      "required_token_positions_per_second": 75939.52090195438,
+      "required_pages_per_second_upper_estimate": 0.005253672259961865
     },
     {
       "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": 9576.770154440801,
-      "required_pages_per_second_upper_estimate": 0.004009593677962406
+      "required_token_positions_per_second": 9492.440112744298,
+      "required_pages_per_second_upper_estimate": 0.003974286450515603
     },
     {
       "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": 19153.540308881602,
-      "required_pages_per_second_upper_estimate": 0.003989177333514726
+      "required_token_positions_per_second": 18984.880225488596,
+      "required_pages_per_second_upper_estimate": 0.003954049886009468
     },
     {
       "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": 28730.310463322403,
-      "required_pages_per_second_upper_estimate": 0.0017223382436321775
+      "required_token_positions_per_second": 28477.320338232894,
+      "required_pages_per_second_upper_estimate": 0.0017071718719266153
     },
     {
       "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": 28730.310463322403,
-      "required_pages_per_second_upper_estimate": 0.0022336552672429172
+      "required_token_positions_per_second": 28477.320338232894,
+      "required_pages_per_second_upper_estimate": 0.0022139863978031652
     },
     {
       "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": 28730.310463322403,
-      "required_pages_per_second_upper_estimate": 0.36852157441954825
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28477.320338232894,
+      "required_pages_per_second_upper_estimate": 0.36527648873453256
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06069746404679677,
-    "producer_p95_seconds": 0.21822235899890075,
-    "gpu_update_seconds": 0.9623286192920157,
-    "producer_batches_per_second": 16.47515288660192,
-    "gpu_batches_per_second": 1.0391460671051216,
-    "producer_headroom": 15.854511129988493,
+    "producer_batch_seconds": 0.3039208876406576,
+    "producer_p95_seconds": 0.24518545899991295,
+    "gpu_update_seconds": 0.9708778660217032,
+    "producer_batches_per_second": 3.290329953176351,
+    "gpu_batches_per_second": 1.0299956719557615,
+    "producer_headroom": 3.1945085234471464,
     "sustainable": true,
-    "buffer_seconds": 60.62670301539699,
+    "buffer_seconds": 60.1944276933456,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409014.5193426829,
-    "upload_bytes_per_second": 3860622.563149292,
+    "generation_bytes_per_second": 381609.0072136321,
+    "upload_bytes_per_second": 3845954.4406717853,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.7649944802319,
+    "estimated_full_shards_per_hour": 81.88440954500888,
     "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": 1.380750825999712,
-      "bytes": 7212989,
-      "pages": 30,
-      "shards": 3
-    },
     {
       "seconds": 1.7552309799993964,
       "bytes": 9939009,
@@ -562,6 +556,12 @@
       "bytes": 7538765,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.4924008759990102,
+      "bytes": 6022523,
+      "pages": 32,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 816200, "loss": 0.5381384491920471, "lr": 3e-05, "grad_norm": 0.2840808439377888, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119468427744, "tokens_per_second": 151122.2313835048, "data_wait_s": 1.8319999071536586e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 816300, "loss": 0.5327563881874084, "lr": 3e-05, "grad_norm": 0.28440403431419115, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119483113319, "tokens_per_second": 151888.8710059361, "data_wait_s": 2.0921001123497263e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 816400, "loss": 0.4969991147518158, "lr": 3e-05, "grad_norm": 0.27524534196188283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119497719635, "tokens_per_second": 150695.83704297701, "data_wait_s": 1.1690000974340364e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 816500, "loss": 0.5110582709312439, "lr": 3e-05, "grad_norm": 0.2823801772473911, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 119512226158, "tokens_per_second": 149414.69378761944, "data_wait_s": 1.541700112284161e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:677754c7daa8a7ca6392b94c991ba90ed1695bdc401f92296628024aef984d80
-size 349866
+oid sha256:c049e773c6cbfae1bd0de9f76480e1076bacc055fa844aefc795e4c896d6aa68
+size 351984