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": 76328.08910145862,
-      "required_pages_per_second_upper_estimate": 0.005280554309606008
+      "required_token_positions_per_second": 75486.1723239331,
+      "required_pages_per_second_upper_estimate": 0.005222308553420727
     },
     {
       "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": 9541.011137682328,
-      "required_pages_per_second_upper_estimate": 0.003994622124378818
+      "required_token_positions_per_second": 9435.771540491638,
+      "required_pages_per_second_upper_estimate": 0.003950560502687769
     },
     {
       "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": 19082.022275364656,
-      "required_pages_per_second_upper_estimate": 0.003974282013190522
+      "required_token_positions_per_second": 18871.543080983276,
+      "required_pages_per_second_upper_estimate": 0.0039304447477105054
     },
     {
       "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": 28623.033413046982,
-      "required_pages_per_second_upper_estimate": 0.0017159071482707373
+      "required_token_positions_per_second": 28307.31462147491,
+      "required_pages_per_second_upper_estimate": 0.0016969802887906733
     },
     {
       "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": 28623.033413046982,
-      "required_pages_per_second_upper_estimate": 0.0022253149484458795
+      "required_token_positions_per_second": 28307.31462147491,
+      "required_pages_per_second_upper_estimate": 0.0022007692011013526
     },
     {
       "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": 28623.033413046982,
-      "required_pages_per_second_upper_estimate": 0.3671455396037375
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28307.31462147491,
+      "required_pages_per_second_upper_estimate": 0.36309583793787803
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.06252847592982391,
-    "producer_p95_seconds": 0.21123036000062712,
-    "gpu_update_seconds": 0.9659353570609835,
-    "producer_batches_per_second": 15.992713481811169,
-    "gpu_batches_per_second": 1.035265965460322,
-    "producer_headroom": 15.447927407427276,
+    "producer_batch_seconds": 0.11905421234381208,
+    "producer_p95_seconds": 0.2112942550011212,
+    "gpu_update_seconds": 0.9767086835932246,
+    "producer_batches_per_second": 8.399534802784958,
+    "gpu_batches_per_second": 1.0238467383345962,
+    "producer_headroom": 8.203898580023571,
     "sustainable": true,
-    "buffer_seconds": 60.85392749484196,
+    "buffer_seconds": 60.55593838277993,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 338271.04925983597,
-    "upload_bytes_per_second": 4124460.623959151,
+    "generation_bytes_per_second": 337829.4984102271,
+    "upload_bytes_per_second": 4061704.1789343897,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.58509262415227,
+    "estimated_full_shards_per_hour": 72.49034609060392,
     "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": [
@@ -448,6 +448,12 @@
       "bytes": 11751236,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.3820675099996151,
+      "bytes": 1056686,
+      "pages": 32,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -95,3 +95,4 @@
 {"step": 809500, "loss": 0.5444309711456299, "lr": 3e-05, "grad_norm": 0.2800186850872327, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118487430031, "tokens_per_second": 151739.90095548856, "data_wait_s": 1.8564998754300177e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809600, "loss": 0.4876198172569275, "lr": 3e-05, "grad_norm": 0.30805431226618685, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118502218881, "tokens_per_second": 153219.5542393025, "data_wait_s": 1.749500006553717e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 809700, "loss": 0.5788338780403137, "lr": 3e-05, "grad_norm": 0.3255125200327241, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118516943629, "tokens_per_second": 152159.41712562114, "data_wait_s": 1.7636999473324977e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 809800, "loss": 0.5648759603500366, "lr": 3e-05, "grad_norm": 0.34168568017784234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 118531461221, "tokens_per_second": 150707.84311820107, "data_wait_s": 7.571001333417371e-06, "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:cbf34fa93f1fe6067eb280b3a9b414e6ee7cda3bc7f44d1bd05aa033e9975699
-size 206394
+oid sha256:d4aa820f5273dc2db822b14f4c6ac8f2164758e4d5bed14d3d0b4069d539985c
+size 208512