Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77497.88376651732,
-      "required_pages_per_second_upper_estimate": 0.005361483418832349
+      "required_token_positions_per_second": 76943.01952991771,
+      "required_pages_per_second_upper_estimate": 0.005323096623481982
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9687.235470814665,
-      "required_pages_per_second_upper_estimate": 0.0040558431991500245
+      "required_token_positions_per_second": 9617.877441239714,
+      "required_pages_per_second_upper_estimate": 0.004026804440527349
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19374.47094162933,
-      "required_pages_per_second_upper_estimate": 0.004035191357983489
+      "required_token_positions_per_second": 19235.75488247943,
+      "required_pages_per_second_upper_estimate": 0.004006300461051046
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29061.706412443997,
-      "required_pages_per_second_upper_estimate": 0.0017422049247697018
+      "required_token_positions_per_second": 28853.632323719143,
+      "required_pages_per_second_upper_estimate": 0.0017297312008613841
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29061.706412443997,
-      "required_pages_per_second_upper_estimate": 0.0022594198446303907
+      "required_token_positions_per_second": 28853.632323719143,
+      "required_pages_per_second_upper_estimate": 0.0022432429994532262
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29061.706412443997,
-      "required_pages_per_second_upper_estimate": 0.37277236582963275
+      "required_token_positions_per_second": 28853.632323719143,
+      "required_pages_per_second_upper_estimate": 0.3701034148320204
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09930809585942768,
-    "producer_p95_seconds": 0.19761427699995693,
-    "gpu_update_seconds": 0.9513550101848576,
-    "producer_batches_per_second": 10.069672480836983,
-    "gpu_batches_per_second": 1.0511323210519385,
-    "producer_headroom": 9.579833365564847,
+    "producer_batch_seconds": 0.8214605102421473,
+    "producer_p95_seconds": 0.19235912399926747,
+    "gpu_update_seconds": 0.9582155788847405,
+    "producer_batches_per_second": 1.2173439715382344,
+    "gpu_batches_per_second": 1.0436064931900733,
+    "producer_headroom": 1.1664779583893583,
     "sustainable": true,
-    "buffer_seconds": 60.886720651830885,
+    "buffer_seconds": 60.36758146973865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 312032.4972492839,
-    "upload_bytes_per_second": 4426883.791479649,
+    "generation_bytes_per_second": 259275.64765156538,
+    "upload_bytes_per_second": 4451515.826357642,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.95491016491782,
+    "estimated_full_shards_per_hour": 55.63451835785123,
     "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": [
@@ -514,6 +514,12 @@
       "bytes": 1124581,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.1323460120001982,
+      "bytes": 7094094,
+      "pages": 25,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 533500, "loss": 0.6668593287467957, "lr": 0.0003, "grad_norm": 0.2902200800299181, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78087696348, "tokens_per_second": 154199.5834391522, "data_wait_s": 8.674000127939507e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 533600, "loss": 0.5504591464996338, "lr": 0.0003, "grad_norm": 0.17225628327564876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78102309957, "tokens_per_second": 153145.91271492338, "data_wait_s": 9.602999853086658e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 533700, "loss": 0.5592876672744751, "lr": 0.0003, "grad_norm": 0.16260922391123797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78116873994, "tokens_per_second": 105166.14067153286, "data_wait_s": 1.436300044588279e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 533800, "loss": 0.5632269978523254, "lr": 0.0003, "grad_norm": 0.15193763187468254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 78131356452, "tokens_per_second": 151070.21003535716, "data_wait_s": 1.2249000064912252e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14ece8c80e263dd4e1b051cd2b1ab098a0d9f0a114cfb78bed5315616c015c54
-size 185214
+oid sha256:7d7f4bd458ec2bbab1e951e383c6fc80c213a74b15a96f2094c027551bdc2b50
+size 187332