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": 76375.43340313055,
-      "required_pages_per_second_upper_estimate": 0.006620028995617115
+      "required_token_positions_per_second": 76157.26651153331,
+      "required_pages_per_second_upper_estimate": 0.006601118842392398
     },
     {
       "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": 9546.929175391318,
-      "required_pages_per_second_upper_estimate": 0.0038084127873748677
+      "required_token_positions_per_second": 9519.658313941663,
+      "required_pages_per_second_upper_estimate": 0.0037975340330068867
     },
     {
       "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": 19093.858350782637,
-      "required_pages_per_second_upper_estimate": 0.004417028089023795
+      "required_token_positions_per_second": 19039.316627883327,
+      "required_pages_per_second_upper_estimate": 0.0044044108213325784
     },
     {
       "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": 28640.787526173957,
-      "required_pages_per_second_upper_estimate": 0.0017440736937450589
+      "required_token_positions_per_second": 28558.97494182499,
+      "required_pages_per_second_upper_estimate": 0.0017390917365956635
     },
     {
       "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": 28640.787526173957,
-      "required_pages_per_second_upper_estimate": 0.002204847106539067
+      "required_token_positions_per_second": 28558.97494182499,
+      "required_pages_per_second_upper_estimate": 0.002198548947324155
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28640.787526173957,
-      "required_pages_per_second_upper_estimate": 0.3556227265253729
+      "required_token_positions_per_second": 28558.97494182499,
+      "required_pages_per_second_upper_estimate": 0.3546068880368649
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.5055529695468408,
-    "producer_p95_seconds": 0.33400089199858485,
-    "gpu_update_seconds": 0.9653365842239262,
-    "producer_batches_per_second": 1.9780320960162956,
-    "gpu_batches_per_second": 1.0359081136492316,
-    "producer_headroom": 1.9094667470536641,
+    "producer_batch_seconds": 0.11405587779671578,
+    "producer_p95_seconds": 0.31318423799893935,
+    "gpu_update_seconds": 0.968101973418841,
+    "producer_batches_per_second": 8.767632315997965,
+    "gpu_batches_per_second": 1.0329490358009616,
+    "producer_headroom": 8.487962147328433,
     "sustainable": true,
-    "buffer_seconds": 60.81620480610735,
+    "buffer_seconds": 60.02232235196814,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 283448.8065840012,
-    "upload_bytes_per_second": 4514531.554258477,
+    "generation_bytes_per_second": 288486.1570566524,
+    "upload_bytes_per_second": 4462140.27604122,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.821515542412065,
+    "estimated_full_shards_per_hour": 61.90241369032554,
     "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": 1.2709916069979954,
-      "bytes": 5895393,
-      "pages": 35,
-      "shards": 2
-    },
-    {
-      "seconds": 1.2639917810010957,
-      "bytes": 13243059,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.182673250001244,
       "bytes": 5984869,
@@ -562,6 +550,18 @@
       "bytes": 828896,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.1922220630003721,
+      "bytes": 10349010,
+      "pages": 34,
+      "shards": 3
+    },
+    {
+      "seconds": 0.9792584309980157,
+      "bytes": 2231089,
+      "pages": 24,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 184800, "loss": 0.6112805008888245, "lr": 0.0003, "grad_norm": 0.12806757929661203, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27046955426, "tokens_per_second": 149143.50961878238, "data_wait_s": 8.838000212563202e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 184900, "loss": 0.5951752122491598, "lr": 0.0003, "grad_norm": 0.138714350567583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27061642898, "tokens_per_second": 129181.3069114033, "data_wait_s": 8.664999768370762e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 185000, "loss": 0.5817938577383757, "lr": 0.0003, "grad_norm": 0.151997777867797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27076235668, "tokens_per_second": 149048.55473883892, "data_wait_s": 9.453000529902056e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 185100, "loss": 0.616595970466733, "lr": 0.0003, "grad_norm": 0.13262788098500142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27090932419, "tokens_per_second": 130173.93343251415, "data_wait_s": 8.660997991682962e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b03dd3ced4b5c4bc38df8e73fe826c2b2152445ecea43f5be6bfcab4da6db39
-size 640263
+oid sha256:9e0c154f6a79b22fbcfe3c05da9999492fedec88369fae98c411eb53926bdf95
+size 642381