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": 75311.59975883062,
-      "required_pages_per_second_upper_estimate": 0.006527818591590918
+      "required_token_positions_per_second": 73616.8829584658,
+      "required_pages_per_second_upper_estimate": 0.006380924834555757
     },
     {
       "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": 9413.949969853827,
-      "required_pages_per_second_upper_estimate": 0.0037553653940696615
+      "required_token_positions_per_second": 9202.110369808224,
+      "required_pages_per_second_upper_estimate": 0.0036708594103271996
     },
     {
       "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": 18827.899939707655,
-      "required_pages_per_second_upper_estimate": 0.004355503291329767
+      "required_token_positions_per_second": 18404.22073961645,
+      "required_pages_per_second_upper_estimate": 0.004257492564887917
     },
     {
       "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": 28241.84990956148,
-      "required_pages_per_second_upper_estimate": 0.0017197804859572729
+      "required_token_positions_per_second": 27606.33110942467,
+      "required_pages_per_second_upper_estimate": 0.0016810807253383965
     },
     {
       "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": 28241.84990956148,
-      "required_pages_per_second_upper_estimate": 0.0021741357844822355
+      "required_token_positions_per_second": 27606.33110942467,
+      "required_pages_per_second_upper_estimate": 0.0021252117880190694
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28241.84990956148,
-      "required_pages_per_second_upper_estimate": 0.3506692564853605
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27606.33110942467,
+      "required_pages_per_second_upper_estimate": 0.3427782399322631
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.23900702490625747,
-    "producer_p95_seconds": 0.15273408399980326,
-    "gpu_update_seconds": 0.9789726979123833,
-    "producer_batches_per_second": 4.1839774391242965,
-    "gpu_batches_per_second": 1.0214789463817087,
-    "producer_headroom": 4.095999681584057,
+    "producer_batch_seconds": 0.06760878265625792,
+    "producer_p95_seconds": 0.15546596600006524,
+    "gpu_update_seconds": 1.0015093961747457,
+    "producer_batches_per_second": 14.790977750394967,
+    "gpu_batches_per_second": 0.9984928786684271,
+    "producer_headroom": 14.813303195632162,
     "sustainable": true,
-    "buffer_seconds": 60.696307270567765,
+    "buffer_seconds": 60.09056377048474,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 514925.50159691053,
-    "upload_bytes_per_second": 5586172.257269365,
+    "generation_bytes_per_second": 513803.9063750795,
+    "upload_bytes_per_second": 5754294.188840838,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.49102579050529,
+    "estimated_full_shards_per_hour": 110.25035756530083,
     "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": [
@@ -298,6 +298,12 @@
       "bytes": 13444268,
       "pages": 40,
       "shards": 5
+    },
+    {
+      "seconds": 1.0623180309999043,
+      "bytes": 9531595,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -21,3 +21,4 @@
 {"step": 84000, "loss": 0.6894025504589081, "lr": 0.0003, "grad_norm": 0.16373860687487832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12292965350, "tokens_per_second": 148794.04983287596, "data_wait_s": 1.3869000213162508e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84100, "loss": 0.575191205367446, "lr": 0.0003, "grad_norm": 0.13894145812826192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12307744461, "tokens_per_second": 149439.80352655245, "data_wait_s": 1.443900009689969e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 84200, "loss": 0.5921671111136675, "lr": 0.0003, "grad_norm": 0.1334533288816832, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12322412062, "tokens_per_second": 148253.383593555, "data_wait_s": 1.087200007532374e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 84300, "loss": 0.650002021342516, "lr": 0.0003, "grad_norm": 0.16830847416344066, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 12336900344, "tokens_per_second": 146661.90606472787, "data_wait_s": 1.3800000033370452e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9f49ee716a5f4bbc02d1facfa585d88cd39e32259350de5366f3b6ea47819ef6
-size 48879
+oid sha256:42579d568a8cf836eb8d324eb7a8e108493d4a7d39e6c7cd0af9c4419a43debd
+size 50997