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": 73498.86629600402,
-      "required_pages_per_second_upper_estimate": 0.006370695449364067
+      "required_token_positions_per_second": 75093.59613953083,
+      "required_pages_per_second_upper_estimate": 0.006508922590395133
     },
     {
       "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": 9187.358287000503,
-      "required_pages_per_second_upper_estimate": 0.0036649745839319064
+      "required_token_positions_per_second": 9386.699517441353,
+      "required_pages_per_second_upper_estimate": 0.003744494781171754
     },
     {
       "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": 18374.716574001006,
-      "required_pages_per_second_upper_estimate": 0.004250667295428365
+      "required_token_positions_per_second": 18773.399034882706,
+      "required_pages_per_second_upper_estimate": 0.0043428954661816795
     },
     {
       "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": 27562.074861001507,
-      "required_pages_per_second_upper_estimate": 0.0016783857520040163
+      "required_token_positions_per_second": 28160.09855232406,
+      "required_pages_per_second_upper_estimate": 0.001714802257217737
     },
     {
       "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": 27562.074861001507,
-      "required_pages_per_second_upper_estimate": 0.0021218048195063175
+      "required_token_positions_per_second": 28160.09855232406,
+      "required_pages_per_second_upper_estimate": 0.002167842338699858
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27562.074861001507,
-      "required_pages_per_second_upper_estimate": 0.34222872544298283
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28160.09855232406,
+      "required_pages_per_second_upper_estimate": 0.349654178232664
     }
   ],
   "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": 60,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10529593931255476,
-    "producer_p95_seconds": 0.3304917559980822,
-    "gpu_update_seconds": 1.0031175134467134,
-    "producer_batches_per_second": 9.497042398108574,
-    "gpu_batches_per_second": 0.9968921752387697,
-    "producer_headroom": 9.526649555488685,
+    "producer_batch_seconds": 0.08295379150749227,
+    "producer_p95_seconds": 0.28749901300034253,
+    "gpu_update_seconds": 0.9818147457341978,
+    "producer_batches_per_second": 12.054904083675083,
+    "gpu_batches_per_second": 1.0185220830557025,
+    "producer_headroom": 11.835682587763594,
     "sustainable": true,
-    "buffer_seconds": 60.1870508068028,
+    "buffer_seconds": 60.87251423552026,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460208.10097406345,
-    "upload_bytes_per_second": 5715564.254599713,
+    "generation_bytes_per_second": 460747.4072104429,
+    "upload_bytes_per_second": 5763301.597659959,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.74994537273815,
+    "estimated_full_shards_per_hour": 98.86566793665853,
     "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": 0.9958830260002287,
-      "bytes": 6097937,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.0204584569983126,
       "bytes": 2643615,
@@ -562,6 +556,12 @@
       "bytes": 7383476,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.2710551140007738,
+      "bytes": 11134870,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -259,3 +259,4 @@
 {"step": 25900, "loss": 0.6570776477456093, "lr": 0.0003, "grad_norm": 0.14541464004758917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3789806964, "tokens_per_second": 148410.55718301475, "data_wait_s": 1.1101001291535795e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 26000, "loss": 0.6907591354101896, "lr": 0.0003, "grad_norm": 0.14134570257389986, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3804476685, "tokens_per_second": 149261.4004277892, "data_wait_s": 1.2415999663062394e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 26100, "loss": 0.6592318769544363, "lr": 0.0003, "grad_norm": 0.1358972821967768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3819172841, "tokens_per_second": 149150.48024003956, "data_wait_s": 1.3214998034527525e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 26200, "loss": 0.6582708470523357, "lr": 0.0003, "grad_norm": 0.13743094417910706, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 3833948129, "tokens_per_second": 150819.41961374847, "data_wait_s": 9.823997970670462e-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 sha256:242174c771e92b6cf3a17e590b873610ce93e6654ba87061841a823db3b32d0e
-size 551104
+oid sha256:5903a87105724683b4e5f786a7fc53b55594180cbf782a394c26bae76640567a
+size 553222