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": 75290.00046203609,
-      "required_pages_per_second_upper_estimate": 0.006525946419287685
+      "required_token_positions_per_second": 74092.57908845434,
+      "required_pages_per_second_upper_estimate": 0.006422156969462345
     },
     {
       "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": 9411.25005775451,
-      "required_pages_per_second_upper_estimate": 0.0037542883587659607
+      "required_token_positions_per_second": 9261.572386056792,
+      "required_pages_per_second_upper_estimate": 0.003694579697645122
     },
     {
       "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": 18822.50011550902,
-      "required_pages_per_second_upper_estimate": 0.004354254137035073
+      "required_token_positions_per_second": 18523.144772113585,
+      "required_pages_per_second_upper_estimate": 0.004285003546814642
     },
     {
       "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": 28233.750173263532,
-      "required_pages_per_second_upper_estimate": 0.00171928725451276
+      "required_token_positions_per_second": 27784.717158170377,
+      "required_pages_per_second_upper_estimate": 0.0016919434997877437
     },
     {
       "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": 28233.750173263532,
-      "required_pages_per_second_upper_estimate": 0.0021735122443605143
+      "required_token_positions_per_second": 27784.717158170377,
+      "required_pages_per_second_upper_estimate": 0.0021389444398557016
     },
     {
       "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": 28233.750173263532,
-      "required_pages_per_second_upper_estimate": 0.3505686848686136
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27784.717158170377,
+      "required_pages_per_second_upper_estimate": 0.34499319763798475
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09228926942984117,
-    "producer_p95_seconds": 0.3614122170001792,
-    "gpu_update_seconds": 0.9792535469192393,
-    "producer_batches_per_second": 10.835495894354281,
-    "gpu_batches_per_second": 1.0211859871695432,
-    "producer_headroom": 10.610697787175283,
+    "producer_batch_seconds": 0.13756458710159336,
+    "producer_p95_seconds": 0.37749076899854117,
+    "gpu_update_seconds": 0.9950794115559253,
+    "producer_batches_per_second": 7.2693126993612545,
+    "gpu_batches_per_second": 1.004944920362065,
+    "producer_headroom": 7.233543403296411,
     "sustainable": true,
-    "buffer_seconds": 60.713719908992836,
+    "buffer_seconds": 60.69984410491144,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353638.87769608555,
-    "upload_bytes_per_second": 6424950.050906844,
+    "generation_bytes_per_second": 371794.4792492379,
+    "upload_bytes_per_second": 6343987.385903432,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.88267086183477,
+    "estimated_full_shards_per_hour": 79.7784403143678,
     "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": 1.0118765539991728,
-      "bytes": 6408285,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0685828210007458,
       "bytes": 12919611,
@@ -562,6 +556,12 @@
       "bytes": 9774660,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.0518995679994987,
+      "bytes": 1242140,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -315,3 +315,4 @@
 {"step": 31500, "loss": 0.7267534174025059, "lr": 0.0003, "grad_norm": 0.13736877923239438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4609858831, "tokens_per_second": 149405.9264729884, "data_wait_s": 1.3778000720776618e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 31600, "loss": 0.6758586764335632, "lr": 0.0003, "grad_norm": 0.13791030854855796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4624445857, "tokens_per_second": 149381.69451310334, "data_wait_s": 1.5675999748054892e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 31700, "loss": 0.6630048379302025, "lr": 0.0003, "grad_norm": 0.13459105664915494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4638943903, "tokens_per_second": 147567.51774466137, "data_wait_s": 1.2213000445626676e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 31800, "loss": 0.6621676087379456, "lr": 0.0003, "grad_norm": 0.12920630094284977, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4653412676, "tokens_per_second": 147342.83977307897, "data_wait_s": 9.068000508705154e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a994efdee26e2794259ef11aa69ef40864718f295b59436e25c5e93a7cb0ea97
-size 670495
+oid sha256:b23979765aaf1b38562b626341eff033ceeba2047dec78f3e5aa3869b63705fd
+size 672613