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": 76641.46286933818,
-      "required_pages_per_second_upper_estimate": 0.005302234233995237
+      "required_token_positions_per_second": 77499.172038135,
+      "required_pages_per_second_upper_estimate": 0.005361572544452063
     },
     {
       "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": 9580.182858667273,
-      "required_pages_per_second_upper_estimate": 0.004011022505956662
+      "required_token_positions_per_second": 9687.396504766875,
+      "required_pages_per_second_upper_estimate": 0.004055910620703044
     },
     {
       "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": 19160.365717334545,
-      "required_pages_per_second_upper_estimate": 0.003990598886097349
+      "required_token_positions_per_second": 19374.79300953375,
+      "required_pages_per_second_upper_estimate": 0.004035258436234477
     },
     {
       "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": 28740.548576001816,
-      "required_pages_per_second_upper_estimate": 0.0017229520028546154
+      "required_token_positions_per_second": 29062.189514300626,
+      "required_pages_per_second_upper_estimate": 0.001742233885988353
     },
     {
       "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": 28740.548576001816,
-      "required_pages_per_second_upper_estimate": 0.0022344512354710434
+      "required_token_positions_per_second": 29062.189514300626,
+      "required_pages_per_second_upper_estimate": 0.0022594574036748035
     },
     {
       "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": 28740.548576001816,
-      "required_pages_per_second_upper_estimate": 0.36865289793617084
+      "required_token_positions_per_second": 29062.189514300626,
+      "required_pages_per_second_upper_estimate": 0.3727785625415352
     }
   ],
   "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": 0.07647222365622497,
-    "producer_p95_seconds": 0.2932725200007553,
-    "gpu_update_seconds": 0.96198581341923,
-    "producer_batches_per_second": 13.076643416247753,
-    "gpu_batches_per_second": 1.0395163692130287,
-    "producer_headroom": 12.579545453572313,
+    "producer_batch_seconds": 0.09684564208586721,
+    "producer_p95_seconds": 0.29432975799863925,
+    "gpu_update_seconds": 0.9513391957751585,
+    "producer_batches_per_second": 10.325709845708495,
+    "gpu_batches_per_second": 1.0511497943540447,
+    "producer_headroom": 9.823252500423955,
     "sustainable": true,
-    "buffer_seconds": 60.605106245411484,
+    "buffer_seconds": 60.88570852961014,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459361.2747683335,
+    "generation_bytes_per_second": 459398.4150982266,
     "upload_bytes_per_second": 5400526.091589437,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.56823618209366,
+    "estimated_full_shards_per_hour": 98.57620563230608,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -278,3 +278,4 @@
 {"step": 672800, "loss": 0.5490952730178833, "lr": 0.0001, "grad_norm": 0.2002931100371676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98480494479, "tokens_per_second": 150418.87794062702, "data_wait_s": 8.977996913017705e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 672900, "loss": 0.5461544394493103, "lr": 0.0001, "grad_norm": 0.19277019882890248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98495069753, "tokens_per_second": 152051.37842661559, "data_wait_s": 9.282000974053517e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 673000, "loss": 0.6332626938819885, "lr": 0.0001, "grad_norm": 0.2485042289402885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98509657315, "tokens_per_second": 151974.2085766073, "data_wait_s": 1.848900137702003e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 673100, "loss": 0.5786046385765076, "lr": 0.0001, "grad_norm": 0.20524825609338687, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98524328627, "tokens_per_second": 152394.9734899505, "data_wait_s": 1.2312997569097206e-05, "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:f4a1b2bedaee7ad7d4bd9083af81e87c5fd94ce126dd666051b7bf087ba43e3e
-size 597120
+oid sha256:05acdbf38c2d0d6fcea3f5237e1e3d98413836d944a31e7712496af6390a5be5
+size 599238