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": 75288.02037586171,
-      "required_pages_per_second_upper_estimate": 0.005208599941877284
+      "required_token_positions_per_second": 77158.93782757637,
+      "required_pages_per_second_upper_estimate": 0.005338034352313482
     },
     {
       "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": 9411.002546982714,
-      "required_pages_per_second_upper_estimate": 0.003940190242340983
+      "required_token_positions_per_second": 9644.867228447047,
+      "required_pages_per_second_upper_estimate": 0.004038104500820212
     },
     {
       "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": 18822.00509396543,
-      "required_pages_per_second_upper_estimate": 0.0039201272914192575
+      "required_token_positions_per_second": 19289.734456894093,
+      "required_pages_per_second_upper_estimate": 0.004017542982864515
     },
     {
       "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": 28233.00764094814,
-      "required_pages_per_second_upper_estimate": 0.001692525698768283
+      "required_token_positions_per_second": 28934.60168534114,
+      "required_pages_per_second_upper_estimate": 0.0017345851904575662
     },
     {
       "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": 28233.00764094814,
-      "required_pages_per_second_upper_estimate": 0.002194992160207262
+      "required_token_positions_per_second": 28934.60168534114,
+      "required_pages_per_second_upper_estimate": 0.0022495380111727534
     },
     {
       "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": 28233.00764094814,
-      "required_pages_per_second_upper_estimate": 0.3621427077763002
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28934.60168534114,
+      "required_pages_per_second_upper_estimate": 0.37114200286478033
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12562474191435058,
-    "producer_p95_seconds": 0.33187363899924094,
-    "gpu_update_seconds": 0.9792793014336997,
-    "producer_batches_per_second": 7.960215358546073,
-    "gpu_batches_per_second": 1.0211591305319785,
-    "producer_headroom": 7.795274135578805,
+    "producer_batch_seconds": 0.10719277659421778,
+    "producer_p95_seconds": 0.3240122749994043,
+    "gpu_update_seconds": 0.9555341490671718,
+    "producer_batches_per_second": 9.32898681956469,
+    "gpu_batches_per_second": 1.0465350725311464,
+    "producer_headroom": 8.914165482291608,
     "sustainable": true,
-    "buffer_seconds": 60.71531668888938,
+    "buffer_seconds": 60.19865139123183,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419236.6621758823,
-    "upload_bytes_per_second": 5153158.690935891,
+    "generation_bytes_per_second": 419614.1421293825,
+    "upload_bytes_per_second": 5076550.506683228,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.95842837293007,
+    "estimated_full_shards_per_hour": 90.03942678366764,
     "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": 0.971025193000969,
-      "bytes": 6720640,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.2639794309980061,
       "bytes": 7601943,
@@ -562,6 +556,12 @@
       "bytes": 6340288,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3758785059981165,
+      "bytes": 2329519,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -339,3 +339,4 @@
 {"step": 638300, "loss": 0.6117377877235413, "lr": 0.0001, "grad_norm": 0.21144039233800835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93430705861, "tokens_per_second": 151678.49098972362, "data_wait_s": 1.4162003935780376e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638400, "loss": 0.5426754355430603, "lr": 0.0001, "grad_norm": 0.16570657956578666, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93445350962, "tokens_per_second": 152011.60157106334, "data_wait_s": 1.5346005966421217e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638500, "loss": 0.521386444568634, "lr": 0.0001, "grad_norm": 0.16600122590102329, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93460130845, "tokens_per_second": 153391.30282328484, "data_wait_s": 9.71899862634018e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 638600, "loss": 0.5342638492584229, "lr": 0.0001, "grad_norm": 0.17172977449817925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93474774785, "tokens_per_second": 152039.84697366503, "data_wait_s": 1.740399602567777e-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:d97d1170e677e5c0f83abcf5aaaed39e8ded9e18c5476035ad91600e35226eb3
-size 727884
+oid sha256:9c9dc49f4c9ddf43e64c04934022192d60e20dfbcf269099c33e2579b421294c
+size 730002