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": 76062.0878954498,
-      "required_pages_per_second_upper_estimate": 0.005262151728966486
+      "required_token_positions_per_second": 76765.8043380277,
+      "required_pages_per_second_upper_estimate": 0.005310836465311134
     },
     {
       "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": 9507.760986931225,
-      "required_pages_per_second_upper_estimate": 0.0039807009806014345
+      "required_token_positions_per_second": 9595.725542253462,
+      "required_pages_per_second_upper_estimate": 0.0040175299030060585
     },
     {
       "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": 19015.52197386245,
-      "required_pages_per_second_upper_estimate": 0.003960431754118496
+      "required_token_positions_per_second": 19191.451084506924,
+      "required_pages_per_second_upper_estimate": 0.003997073148303096
     },
     {
       "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": 28523.282960793676,
-      "required_pages_per_second_upper_estimate": 0.0017099272609682732
+      "required_token_positions_per_second": 28787.176626760385,
+      "required_pages_per_second_upper_estimate": 0.0017257472833006806
     },
     {
       "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": 28523.282960793676,
-      "required_pages_per_second_upper_estimate": 0.0022175597895390553
+      "required_token_positions_per_second": 28787.176626760385,
+      "required_pages_per_second_upper_estimate": 0.002238076361322402
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28523.282960793676,
-      "required_pages_per_second_upper_estimate": 0.3658660479059232
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28787.176626760385,
+      "required_pages_per_second_upper_estimate": 0.3692509925060015
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.7422414471641332,
-    "producer_p95_seconds": 0.2903245550005522,
-    "gpu_update_seconds": 0.969313386471088,
-    "producer_batches_per_second": 1.3472704923993124,
-    "gpu_batches_per_second": 1.0316580932000028,
-    "producer_headroom": 1.3059273234801478,
+    "producer_batch_seconds": 0.1074779531327863,
+    "producer_p95_seconds": 0.2930474429995229,
+    "gpu_update_seconds": 0.9604276361822364,
+    "producer_batches_per_second": 9.304233760057985,
+    "gpu_batches_per_second": 1.0412028583174329,
+    "producer_headroom": 8.93604323665945,
     "sustainable": true,
-    "buffer_seconds": 60.09742996120745,
+    "buffer_seconds": 60.50694107948089,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 288139.03597956366,
-    "upload_bytes_per_second": 5913793.068473682,
+    "generation_bytes_per_second": 287174.4858517668,
+    "upload_bytes_per_second": 6053612.027126391,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.827929587747406,
+    "estimated_full_shards_per_hour": 61.620959583899996,
     "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": 0.8807951420021709,
-      "bytes": 343455,
-      "pages": 9,
-      "shards": 1
-    },
     {
       "seconds": 1.1079528180016496,
       "bytes": 5894448,
@@ -562,6 +556,12 @@
       "bytes": 6305836,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.0651401020004414,
+      "bytes": 11486452,
+      "pages": 47,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -307,3 +307,4 @@
 {"step": 427800, "loss": 0.5700190663337708, "lr": 0.0003, "grad_norm": 0.1584146694789872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62617848647, "tokens_per_second": 150644.28702120512, "data_wait_s": 1.571300163050182e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427900, "loss": 0.5963143706321716, "lr": 0.0003, "grad_norm": 0.14141060153501045, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62632384124, "tokens_per_second": 120423.55888231561, "data_wait_s": 7.5739990279544145e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 428000, "loss": 0.5572988986968994, "lr": 0.0003, "grad_norm": 0.16986066606522784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62647100625, "tokens_per_second": 153253.55717507965, "data_wait_s": 9.002000297186896e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 428100, "loss": 0.6150058507919312, "lr": 0.0003, "grad_norm": 0.199323124422195, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62661874468, "tokens_per_second": 152834.9155989397, "data_wait_s": 1.4957000530557707e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26e20017ab7d89292724f23cb797650ababc412ac4be88a504331803b7f76925
-size 659325
+oid sha256:1acee676b40c1f01ca3e9816787b312f6cee86e15b36da35873e283d862909c4
+size 661443