Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74637.90713111241,
-      "required_pages_per_second_upper_estimate": 0.006469424622078658
+      "required_token_positions_per_second": 74397.70421670156,
+      "required_pages_per_second_upper_estimate": 0.006448604442246247
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9329.738391389052,
-      "required_pages_per_second_upper_estimate": 0.003721772136344763
+      "required_token_positions_per_second": 9299.713027087695,
+      "required_pages_per_second_upper_estimate": 0.0037097945696057504
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18659.476782778103,
-      "required_pages_per_second_upper_estimate": 0.004316541558120439
+      "required_token_positions_per_second": 18599.42605417539,
+      "required_pages_per_second_upper_estimate": 0.004302649878914928
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27989.215174167155,
-      "required_pages_per_second_upper_estimate": 0.001704396356043777
+      "required_token_positions_per_second": 27899.139081263085,
+      "required_pages_per_second_upper_estimate": 0.0016989111945786548
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27989.215174167155,
-      "required_pages_per_second_upper_estimate": 0.002154687263213873
+      "required_token_positions_per_second": 27899.139081263085,
+      "required_pages_per_second_upper_estimate": 0.0021477529562355336
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27989.215174167155,
-      "required_pages_per_second_upper_estimate": 0.347532378585832
+      "required_token_positions_per_second": 27899.139081263085,
+      "required_pages_per_second_upper_estimate": 0.34641393497725376
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.45482303443753835,
-    "producer_p95_seconds": 0.3236161329987226,
-    "gpu_update_seconds": 0.9878090481621621,
-    "producer_batches_per_second": 2.198657333256352,
-    "gpu_batches_per_second": 1.0123414053156523,
-    "producer_headroom": 2.1718536075987145,
+    "producer_batch_seconds": 0.39163886153170324,
+    "producer_p95_seconds": 0.3503029629937373,
+    "gpu_update_seconds": 0.9909983214703658,
+    "producer_batches_per_second": 2.553372757976547,
+    "gpu_batches_per_second": 1.0090834447794808,
+    "producer_headroom": 2.5303881172429166,
     "sustainable": true,
-    "buffer_seconds": 60.256351937891885,
+    "buffer_seconds": 60.45089760969231,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294094.5589061511,
-    "upload_bytes_per_second": 5165830.519094724,
+    "generation_bytes_per_second": 282533.83858596196,
+    "upload_bytes_per_second": 5268848.120160666,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.105846170314784,
+    "estimated_full_shards_per_hour": 60.625184709397736,
     "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": 1.0084878679990652,
-      "bytes": 5849421,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.7869264700020722,
       "bytes": 2542185,
@@ -562,6 +556,12 @@
       "bytes": 540912,
       "pages": 18,
       "shards": 1
+    },
+    {
+      "seconds": 1.4004500939990976,
+      "bytes": 16241287,
+      "pages": 47,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -329,3 +329,4 @@
 {"step": 387900, "loss": 0.576649721711874, "lr": 0.0003, "grad_norm": 0.15065807361659975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56778200568, "tokens_per_second": 149313.68857831822, "data_wait_s": 1.7134996596723795e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 388000, "loss": 0.5786723587661982, "lr": 0.0003, "grad_norm": 0.15961557350696104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56792864148, "tokens_per_second": 131251.37765346418, "data_wait_s": 9.183997462969273e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 388100, "loss": 0.6169699393212795, "lr": 0.0003, "grad_norm": 0.15993424197939457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56807507607, "tokens_per_second": 149504.69687023005, "data_wait_s": 1.0605996067170054e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 388200, "loss": 0.5952812060713768, "lr": 0.0003, "grad_norm": 0.16336263963922182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56822193731, "tokens_per_second": 150220.98964047353, "data_wait_s": 1.0478004696778953e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f12b58b128034cb4e5823f856c349227ea47dd31a17ccd896a970951d2c46efc
-size 705921
+oid sha256:66e2b6a8d7d78aa9be7efeb8e45da0c620b178a39905a47e3d521fcd7916f021
+size 708039