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": 73436.5976740253,
-      "required_pages_per_second_upper_estimate": 0.006365298163029331
+      "required_token_positions_per_second": 75798.6672275844,
+      "required_pages_per_second_upper_estimate": 0.006570036365321282
     },
     {
       "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": 9179.574709253162,
-      "required_pages_per_second_upper_estimate": 0.003661869598393634
+      "required_token_positions_per_second": 9474.83340344805,
+      "required_pages_per_second_upper_estimate": 0.003779652706018849
     },
     {
       "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": 18359.149418506324,
-      "required_pages_per_second_upper_estimate": 0.0042470661079772485
+      "required_token_positions_per_second": 18949.6668068961,
+      "required_pages_per_second_upper_estimate": 0.004383671913030145
     },
     {
       "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": 27538.724127759484,
-      "required_pages_per_second_upper_estimate": 0.0016769638148614063
+      "required_token_positions_per_second": 28424.500210344155,
+      "required_pages_per_second_upper_estimate": 0.0017309029309828703
     },
     {
       "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": 27538.724127759484,
-      "required_pages_per_second_upper_estimate": 0.00212000721542238
+      "required_token_positions_per_second": 28424.500210344155,
+      "required_pages_per_second_upper_estimate": 0.002188196710244879
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 27538.724127759484,
-      "required_pages_per_second_upper_estimate": 0.3419387874859938
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28424.500210344155,
+      "required_pages_per_second_upper_estimate": 0.3529371619298478
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.41315269255511566,
-    "producer_p95_seconds": 0.4113726069990662,
-    "gpu_update_seconds": 1.0039680804286195,
-    "producer_batches_per_second": 2.4204126416690297,
-    "gpu_batches_per_second": 0.9960476030005601,
-    "producer_headroom": 2.43001703370162,
+    "producer_batch_seconds": 0.12504075224234157,
+    "producer_p95_seconds": 0.42566588499903446,
+    "gpu_update_seconds": 0.9726820100758862,
+    "producer_batches_per_second": 7.997392706514588,
+    "gpu_batches_per_second": 1.028085221728304,
+    "producer_headroom": 7.7789200131388405,
     "sustainable": true,
-    "buffer_seconds": 60.23808482571717,
+    "buffer_seconds": 60.306284624704944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 324828.29635209485,
-    "upload_bytes_per_second": 5287585.875586291,
+    "generation_bytes_per_second": 326172.4089895161,
+    "upload_bytes_per_second": 5278868.57076614,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.70059078142295,
+    "estimated_full_shards_per_hour": 69.98900606407273,
     "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": 0.9463060489979398,
-      "bytes": 743440,
-      "pages": 25,
-      "shards": 1
-    },
-    {
-      "seconds": 1.0638403249977273,
-      "bytes": 11964464,
-      "pages": 46,
-      "shards": 4
-    },
     {
       "seconds": 1.8820245999959297,
       "bytes": 865291,
@@ -562,6 +550,18 @@
       "bytes": 839375,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.33888140199997,
+      "bytes": 10777111,
+      "pages": 37,
+      "shards": 3
+    },
+    {
+      "seconds": 0.8399469979995047,
+      "bytes": 2135537,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -398,3 +398,4 @@
 {"step": 354800, "loss": 0.5827666632831097, "lr": 0.0003, "grad_norm": 0.14574788928946014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51932766104, "tokens_per_second": 146994.49737078985, "data_wait_s": 9.767994924914092e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 354900, "loss": 0.584769681096077, "lr": 0.0003, "grad_norm": 0.1662222847641233, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51947299203, "tokens_per_second": 140361.33777305816, "data_wait_s": 9.903000318445265e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 355000, "loss": 0.5933996718376875, "lr": 0.0003, "grad_norm": 0.14682717746036786, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51961890047, "tokens_per_second": 147682.00578222997, "data_wait_s": 1.4045996067579836e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 355100, "loss": 0.6090170033276081, "lr": 0.0003, "grad_norm": 0.15771364523541026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51976574318, "tokens_per_second": 116100.11540562313, "data_wait_s": 1.5910001820884645e-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 sha256:0ead4502e42dd8a552f743849460503339702c181edea594ae1124b002f83341
-size 853629
+oid sha256:131b5fb40908cfdd8b136c878a6dd5c0b77d7208056f2240e0397cbb81a8447b
+size 855747