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": 75120.86542345275,
-      "required_pages_per_second_upper_estimate": 0.005197035774416706
+      "required_token_positions_per_second": 75288.02037586171,
+      "required_pages_per_second_upper_estimate": 0.005208599941877284
     },
     {
       "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": 9390.108177931594,
-      "required_pages_per_second_upper_estimate": 0.003931442206343321
+      "required_token_positions_per_second": 9411.002546982714,
+      "required_pages_per_second_upper_estimate": 0.003940190242340983
     },
     {
       "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": 18780.216355863187,
-      "required_pages_per_second_upper_estimate": 0.003911423799315693
+      "required_token_positions_per_second": 18822.00509396543,
+      "required_pages_per_second_upper_estimate": 0.0039201272914192575
     },
     {
       "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": 28170.32453379478,
-      "required_pages_per_second_upper_estimate": 0.001688767942206002
+      "required_token_positions_per_second": 28233.00764094814,
+      "required_pages_per_second_upper_estimate": 0.001692525698768283
     },
     {
       "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": 28170.32453379478,
-      "required_pages_per_second_upper_estimate": 0.0021901188243399383
+      "required_token_positions_per_second": 28233.00764094814,
+      "required_pages_per_second_upper_estimate": 0.002194992160207262
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28170.32453379478,
-      "required_pages_per_second_upper_estimate": 0.36133867618161364
+      "required_token_positions_per_second": 28233.00764094814,
+      "required_pages_per_second_upper_estimate": 0.3621427077763002
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.14094225026542517,
-    "producer_p95_seconds": 0.33261065400438383,
-    "gpu_update_seconds": 0.9814583416258421,
-    "producer_batches_per_second": 7.095104541872864,
-    "gpu_batches_per_second": 1.018891946390147,
-    "producer_headroom": 6.963549537328522,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.85041718080221,
+    "buffer_seconds": 60.71531668888938,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 419194.5397698384,
+    "generation_bytes_per_second": 419236.6621758823,
     "upload_bytes_per_second": 5153158.690935891,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.94938988515247,
+    "estimated_full_shards_per_hour": 89.95842837293007,
     "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": [

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

@@ -338,3 +338,4 @@
 {"step": 638200, "loss": 0.5692286491394043, "lr": 0.0001, "grad_norm": 0.1775209778022932, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93416081425, "tokens_per_second": 153340.83732563278, "data_wait_s": 9.074996341951191e-06, "gpu_peak_gb": 89.37861251831055}
 {"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}

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:a1c58361050a55aed4ab35687eef2c209acf0ee714b8fc6363846cd8f428d459
-size 725766
+oid sha256:d97d1170e677e5c0f83abcf5aaaed39e8ded9e18c5476035ad91600e35226eb3
+size 727884