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": 77479.99404800087,
-      "required_pages_per_second_upper_estimate": 0.005360245766595515
+      "required_token_positions_per_second": 76989.844837656,
+      "required_pages_per_second_upper_estimate": 0.005326336106920999
     },
     {
       "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": 9684.999256000108,
-      "required_pages_per_second_upper_estimate": 0.004054906942704646
+      "required_token_positions_per_second": 9623.730604707,
+      "required_pages_per_second_upper_estimate": 0.0040292550378431494
     },
     {
       "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": 19369.998512000217,
-      "required_pages_per_second_upper_estimate": 0.004034259868837645
+      "required_token_positions_per_second": 19247.461209414,
+      "required_pages_per_second_upper_estimate": 0.004008738580234916
     },
     {
       "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": 29054.99776800033,
-      "required_pages_per_second_upper_estimate": 0.0017418027517788108
+      "required_token_positions_per_second": 28871.191814121,
+      "required_pages_per_second_upper_estimate": 0.0017307838654991863
     },
     {
       "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": 29054.99776800033,
-      "required_pages_per_second_upper_estimate": 0.002258898277033104
+      "required_token_positions_per_second": 28871.191814121,
+      "required_pages_per_second_upper_estimate": 0.002244608172595935
     },
     {
       "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": 29054.99776800033,
-      "required_pages_per_second_upper_estimate": 0.37268631454189055
+      "required_token_positions_per_second": 28871.191814121,
+      "required_pages_per_second_upper_estimate": 0.37032864912098357
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1323857838356446,
-    "producer_p95_seconds": 0.23241904099995736,
-    "gpu_update_seconds": 0.9515746729965362,
-    "producer_batches_per_second": 7.5536811508514266,
-    "gpu_batches_per_second": 1.0508896762152895,
-    "producer_headroom": 7.1878916710415455,
+    "producer_batch_seconds": 0.0631036542266088,
+    "producer_p95_seconds": 0.21637895599997137,
+    "gpu_update_seconds": 0.9576327911228544,
+    "producer_batches_per_second": 15.846942815846186,
+    "gpu_batches_per_second": 1.0442416020732421,
+    "producer_headroom": 15.175552079503047,
     "sustainable": true,
-    "buffer_seconds": 60.90077907177832,
+    "buffer_seconds": 60.330865840739826,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 271334.631729472,
-    "upload_bytes_per_second": 4822793.520367699,
+    "generation_bytes_per_second": 283475.25884655624,
+    "upload_bytes_per_second": 4836205.093851002,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.22209562218781,
+    "estimated_full_shards_per_hour": 60.82719158217922,
     "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": 1.1802944610008126,
-      "bytes": 6153792,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.060665737999443,
       "bytes": 2571654,
@@ -562,6 +556,12 @@
       "bytes": 6099339,
       "pages": 40,
       "shards": 2
+    },
+    {
+      "seconds": 1.2393856140006392,
+      "bytes": 7638444,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 945900, "loss": 0.4961773753166199, "lr": 3e-05, "grad_norm": 0.3491370009812067, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138453339381, "tokens_per_second": 152265.35082931517, "data_wait_s": 8.971001079771668e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 946000, "loss": 0.5332742929458618, "lr": 3e-05, "grad_norm": 0.340665019120442, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138467927763, "tokens_per_second": 152139.04841670825, "data_wait_s": 1.3922999642090872e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 946100, "loss": 0.5372671484947205, "lr": 3e-05, "grad_norm": 0.36609023324004225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138482617984, "tokens_per_second": 153110.65295074024, "data_wait_s": 1.601499752723612e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 946200, "loss": 0.4899270534515381, "lr": 3e-05, "grad_norm": 0.349797071505218, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138497380012, "tokens_per_second": 153382.1746692865, "data_wait_s": 1.4434997865464538e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:635f8e51d245408860a947dac71200748c72e6ec37aefca7810824d7ac5b2a4d
-size 450195
+oid sha256:b47c449eee62660a611f736871068670e193499d7c4af28b2e72e517627492f5
+size 452313