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": 74063.33551674566,
-      "required_pages_per_second_upper_estimate": 0.006419622210783791
+      "required_token_positions_per_second": 75748.67018818353,
+      "required_pages_per_second_upper_estimate": 0.006565702748662353
     },
     {
       "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": 9257.916939593208,
-      "required_pages_per_second_upper_estimate": 0.003693121485397003
+      "required_token_positions_per_second": 9468.583773522942,
+      "required_pages_per_second_upper_estimate": 0.0037771596352014288
     },
     {
       "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": 18515.833879186415,
-      "required_pages_per_second_upper_estimate": 0.004283312300403262
+      "required_token_positions_per_second": 18937.167547045883,
+      "required_pages_per_second_upper_estimate": 0.004380780429243257
     },
     {
       "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": 27773.750818779623,
-      "required_pages_per_second_upper_estimate": 0.001691275707254783
+      "required_token_positions_per_second": 28405.751320568826,
+      "required_pages_per_second_upper_estimate": 0.0017297612219633752
     },
     {
       "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": 27773.750818779623,
-      "required_pages_per_second_upper_estimate": 0.0021381002206926313
+      "required_token_positions_per_second": 28405.751320568826,
+      "required_pages_per_second_upper_estimate": 0.0021867533688097254
     },
     {
       "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": 27773.750818779623,
-      "required_pages_per_second_upper_estimate": 0.34485703240472854
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28405.751320568826,
+      "required_pages_per_second_upper_estimate": 0.35270436346733586
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.45545081864068493,
+    "producer_batch_seconds": 0.17128138023443285,
     "producer_p95_seconds": 0.38295882899910794,
-    "gpu_update_seconds": 0.9954723141429427,
-    "producer_batches_per_second": 2.195626748425985,
-    "gpu_batches_per_second": 1.0045482790357212,
-    "producer_headroom": 2.1856856402497598,
+    "gpu_update_seconds": 0.9733240176604612,
+    "producer_batches_per_second": 5.838346226725286,
+    "gpu_batches_per_second": 1.027407093481222,
+    "producer_headroom": 5.682602605889049,
     "sustainable": true,
-    "buffer_seconds": 60.72381116271951,
+    "buffer_seconds": 60.346089094948596,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353925.6583919374,
+    "generation_bytes_per_second": 354235.1055389954,
     "upload_bytes_per_second": 6320002.38046379,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.94420732325163,
+    "estimated_full_shards_per_hour": 76.01060747744938,
     "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": [

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

@@ -313,3 +313,4 @@
 {"step": 31300, "loss": 0.6205551885068417, "lr": 0.0003, "grad_norm": 0.12686401793089822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4580321298, "tokens_per_second": 148694.87510512723, "data_wait_s": 1.2466000043787062e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 31400, "loss": 0.6879486478865147, "lr": 0.0003, "grad_norm": 0.16777613929339316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4595072175, "tokens_per_second": 150432.5182586362, "data_wait_s": 1.0510000720387325e-05, "gpu_peak_gb": 85.43514251708984}
 {"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}

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:e94967f4ee0c448075694471a7a1d35d1da50cfbe18246dde02eb925d9011ae4
-size 666259
+oid sha256:3072af84a13871c8f627426e9fc17d7b7e95dc596ebbffe2a17d7d48a28a5f1d
+size 668377