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": 75085.65875372152,
-      "required_pages_per_second_upper_estimate": 0.006508234597910304
+      "required_token_positions_per_second": 75798.46097551183,
+      "required_pages_per_second_upper_estimate": 0.0065700184879143725
     },
     {
       "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": 9385.70734421519,
-      "required_pages_per_second_upper_estimate": 0.0037440989884375262
+      "required_token_positions_per_second": 9474.807621938979,
+      "required_pages_per_second_upper_estimate": 0.0037796424213894122
     },
     {
       "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": 18771.41468843038,
-      "required_pages_per_second_upper_estimate": 0.004342436422553239
+      "required_token_positions_per_second": 18949.615243877957,
+      "required_pages_per_second_upper_estimate": 0.004383659984833375
     },
     {
       "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": 28157.122032645573,
-      "required_pages_per_second_upper_estimate": 0.001714621002785909
+      "required_token_positions_per_second": 28424.422865816938,
+      "required_pages_per_second_upper_estimate": 0.001730898221106958
     },
     {
       "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": 28157.122032645573,
-      "required_pages_per_second_upper_estimate": 0.002167613197975474
+      "required_token_positions_per_second": 28424.422865816938,
+      "required_pages_per_second_upper_estimate": 0.00218819075604907
     },
     {
       "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": 28157.122032645573,
-      "required_pages_per_second_upper_estimate": 0.34961721982002775
+      "required_token_positions_per_second": 28424.422865816938,
+      "required_pages_per_second_upper_estimate": 0.35293620156967526
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,39 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.08081432175779923,
-    "producer_p95_seconds": 0.34906973999750335,
-    "gpu_update_seconds": 0.981918534427798,
-    "producer_batches_per_second": 12.374044330867529,
-    "gpu_batches_per_second": 1.0184144253705718,
-    "producer_headroom": 12.150303474310046,
+    "producer_batch_seconds": 0.33070550625009787,
+    "producer_p95_seconds": 0.3365079510003852,
+    "gpu_update_seconds": 0.9726846568008718,
+    "producer_batches_per_second": 3.0238383731166074,
+    "gpu_batches_per_second": 1.0280824242555315,
+    "producer_headroom": 2.9412411901762336,
     "sustainable": true,
-    "buffer_seconds": 60.878949134523474,
+    "buffer_seconds": 60.30644872165405,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 506343.7655790157,
-    "upload_bytes_per_second": 5487182.841082292,
+    "generation_bytes_per_second": 460414.1072867045,
+    "upload_bytes_per_second": 5494127.6962746335,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 108.64958501365523,
+    "estimated_full_shards_per_hour": 98.79414953184938,
     "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": 1.1056661020011234,
-      "bytes": 6106174,
-      "pages": 28,
-      "shards": 2
-    },
-    {
-      "seconds": 1.0782650920009473,
-      "bytes": 5102757,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0556685640003707,
       "bytes": 7468948,
@@ -562,6 +550,18 @@
       "bytes": 7666167,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.7579026999992493,
+      "bytes": 928826,
+      "pages": 27,
+      "shards": 1
+    },
+    {
+      "seconds": 1.3881743410020135,
+      "bytes": 10608461,
+      "pages": 29,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -248,3 +248,4 @@
 {"step": 179800, "loss": 0.5804962553083897, "lr": 0.0003, "grad_norm": 0.15476970584776104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26315108925, "tokens_per_second": 149339.41115634065, "data_wait_s": 1.3672000932274386e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 179900, "loss": 0.6551252659410238, "lr": 0.0003, "grad_norm": 0.1586618530744718, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26329795713, "tokens_per_second": 149333.07682973368, "data_wait_s": 1.1494001228129491e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 180000, "loss": 0.5781124047935009, "lr": 0.0003, "grad_norm": 0.1376944556159123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26344430206, "tokens_per_second": 149420.23540266373, "data_wait_s": 9.59699900704436e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 180100, "loss": 0.6057069599628448, "lr": 0.0003, "grad_norm": 0.14725403066453563, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26359050430, "tokens_per_second": 131940.5248773219, "data_wait_s": 1.0952997399726883e-05, "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:d17241ad13602b43194162012dc2a370eb54a32a612de848b30e30a92ed092a3
-size 533580
+oid sha256:129b791b5b6892b272e7f923122b7fd9b7cf9b88f6fca8a7a9e37e97df2a55ce
+size 535698