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": 77250.23767636965,
-      "required_pages_per_second_upper_estimate": 0.005344350687697837
+      "required_token_positions_per_second": 77536.51844221208,
+      "required_pages_per_second_upper_estimate": 0.005364156255341711
     },
     {
       "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": 9656.279709546207,
-      "required_pages_per_second_upper_estimate": 0.004042882668336725
+      "required_token_positions_per_second": 9692.06480527651,
+      "required_pages_per_second_upper_estimate": 0.004057865140641228
     },
     {
       "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": 19312.559419092413,
-      "required_pages_per_second_upper_estimate": 0.0040222968205556936
+      "required_token_positions_per_second": 19384.12961055302,
+      "required_pages_per_second_upper_estimate": 0.004037203004004065
     },
     {
       "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": 28968.83912863862,
-      "required_pages_per_second_upper_estimate": 0.0017366376729057013
+      "required_token_positions_per_second": 29076.194415829526,
+      "required_pages_per_second_upper_estimate": 0.001743073458450764
     },
     {
       "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": 28968.83912863862,
-      "required_pages_per_second_upper_estimate": 0.002252199821794535
+      "required_token_positions_per_second": 29076.194415829526,
+      "required_pages_per_second_upper_estimate": 0.0022605462231676237
     },
     {
       "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": 28968.83912863862,
-      "required_pages_per_second_upper_estimate": 0.3715811640260979
+      "required_token_positions_per_second": 29076.194415829526,
+      "required_pages_per_second_upper_estimate": 0.37295820238105626
     }
   ],
   "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": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0739088230078977,
-    "producer_p95_seconds": 0.17580693800118752,
-    "gpu_update_seconds": 0.9544048305569539,
-    "producer_batches_per_second": 13.530184344745184,
-    "gpu_batches_per_second": 1.0477734059837465,
-    "producer_headroom": 12.913273296950877,
+    "producer_batch_seconds": 0.33654528545315543,
+    "producer_p95_seconds": 0.19796508400031598,
+    "gpu_update_seconds": 0.9508809717184998,
+    "producer_batches_per_second": 2.9713683216614024,
+    "gpu_batches_per_second": 1.0516563373780934,
+    "producer_headroom": 2.825417597034962,
     "sustainable": true,
-    "buffer_seconds": 60.1275043250881,
+    "buffer_seconds": 60.85638218998399,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 361195.247056976,
-    "upload_bytes_per_second": 3103863.2279518405,
+    "generation_bytes_per_second": 338122.13835619466,
+    "upload_bytes_per_second": 3134139.371303728,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.50409182340584,
+    "estimated_full_shards_per_hour": 72.55313981069926,
     "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": [
@@ -436,6 +436,12 @@
       "bytes": 6065504,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.5038798329987912,
+      "bytes": 7468080,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 653500, "loss": 0.5341771841049194, "lr": 0.0001, "grad_norm": 0.2140702360987653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95655384593, "tokens_per_second": 151370.2923929644, "data_wait_s": 1.1362000805092975e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 653600, "loss": 0.5025205016136169, "lr": 0.0001, "grad_norm": 0.1855704622409398, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95669986827, "tokens_per_second": 152760.85696883049, "data_wait_s": 1.1673999324557371e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 653700, "loss": 0.5879846215248108, "lr": 0.0001, "grad_norm": 0.19696340840845797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95684610244, "tokens_per_second": 153054.70798256705, "data_wait_s": 1.1722000635927543e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 653800, "loss": 0.5489779710769653, "lr": 0.0001, "grad_norm": 0.18465427895519032, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95699267514, "tokens_per_second": 132744.69051038052, "data_wait_s": 9.800000043469481e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ef7ccfd328e8b21b77240230f05a7955bd54e1af3a9bcf22d9b213990154ee28
-size 185214
+oid sha256:c8728ebd70cb9ae820f607129bff2db3339999d9db2d7f37d12a063e5ab554f9
+size 187332