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": 74773.30816650837,
-      "required_pages_per_second_upper_estimate": 0.006481160840655993
+      "required_token_positions_per_second": 72538.34154031264,
+      "required_pages_per_second_upper_estimate": 0.006287439597968475
     },
     {
       "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": 9346.663520813547,
-      "required_pages_per_second_upper_estimate": 0.0037285238235254294
+      "required_token_positions_per_second": 9067.29269253908,
+      "required_pages_per_second_upper_estimate": 0.003617078623160635
     },
     {
       "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": 18693.327041627093,
-      "required_pages_per_second_upper_estimate": 0.004324372219760945
+      "required_token_positions_per_second": 18134.58538507816,
+      "required_pages_per_second_upper_estimate": 0.004195117171035647
     },
     {
       "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": 28039.990562440642,
-      "required_pages_per_second_upper_estimate": 0.0017074883107917585
+      "required_token_positions_per_second": 27201.878077617243,
+      "required_pages_per_second_upper_estimate": 0.0016564516577023851
     },
     {
       "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": 28039.990562440642,
-      "required_pages_per_second_upper_estimate": 0.0021585960931584367
+      "required_token_positions_per_second": 27201.878077617243,
+      "required_pages_per_second_upper_estimate": 0.002094075874032884
     },
     {
       "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": 28039.990562440642,
-      "required_pages_per_second_upper_estimate": 0.34816283897389577
+      "required_token_positions_per_second": 27201.878077617243,
+      "required_pages_per_second_upper_estimate": 0.33775628689443665
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.09649943570332198,
-    "producer_p95_seconds": 0.33812447899981635,
-    "gpu_update_seconds": 0.986020303338985,
-    "producer_batches_per_second": 10.362754898115691,
-    "gpu_batches_per_second": 1.014177899393831,
-    "producer_headroom": 10.217886728067587,
+    "producer_batch_seconds": 0.4623070513593177,
+    "producer_p95_seconds": 0.3272022129967809,
+    "gpu_update_seconds": 1.016400408865513,
+    "producer_batches_per_second": 2.1630645629559577,
+    "gpu_batches_per_second": 0.9838642244508552,
+    "producer_headroom": 2.1985397061909375,
     "sustainable": true,
-    "buffer_seconds": 60.14723850367809,
+    "buffer_seconds": 60.98402453193078,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438545.68296364223,
-    "upload_bytes_per_second": 5666406.5054018255,
+    "generation_bytes_per_second": 436262.4856232367,
+    "upload_bytes_per_second": 5684871.883889803,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.1016947429843,
+    "estimated_full_shards_per_hour": 93.61177374384714,
     "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": 1.1473543519987288,
-      "bytes": 5736316,
-      "pages": 34,
-      "shards": 2
-    },
     {
       "seconds": 1.0078334160007216,
       "bytes": 7473663,
@@ -562,6 +556,12 @@
       "bytes": 7379943,
       "pages": 33,
       "shards": 3
+    },
+    {
+      "seconds": 1.0861735590005992,
+      "bytes": 6754175,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -196,3 +196,4 @@
 {"step": 334600, "loss": 0.5858848206698895, "lr": 0.0003, "grad_norm": 0.16661552085024728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48976990553, "tokens_per_second": 147276.2463025926, "data_wait_s": 1.2868000339949504e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 334700, "loss": 0.6253331005573273, "lr": 0.0003, "grad_norm": 0.14598857780237767, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48991511893, "tokens_per_second": 146692.67291553732, "data_wait_s": 1.7004000255838037e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 334800, "loss": 0.5978746525943279, "lr": 0.0003, "grad_norm": 0.16581356789209187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49006190754, "tokens_per_second": 148435.20168036572, "data_wait_s": 9.647999831940979e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 334900, "loss": 0.6604721676558256, "lr": 0.0003, "grad_norm": 0.1942763048124513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49020784092, "tokens_per_second": 147597.65132997284, "data_wait_s": 1.2881999282399192e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebcf9d3afa2431ea67e1fbf2faa240546cc02f6057d5ddc1e28394eac7f3acbb
-size 421878
+oid sha256:c6723c0a12af8d81183a653b01fe2a066744f07591e1f5d54327b7183a51c20b
+size 423996