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": 77402.66430923995,
-      "required_pages_per_second_upper_estimate": 0.00535489591583832
+      "required_token_positions_per_second": 77268.39899607733,
+      "required_pages_per_second_upper_estimate": 0.005345607129935282
     },
     {
       "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": 9675.333038654993,
-      "required_pages_per_second_upper_estimate": 0.004050859899355818
+      "required_token_positions_per_second": 9658.549874509667,
+      "required_pages_per_second_upper_estimate": 0.00404383313900049
     },
     {
       "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": 19350.666077309987,
-      "required_pages_per_second_upper_estimate": 0.004030233432522258
+      "required_token_positions_per_second": 19317.099749019333,
+      "required_pages_per_second_upper_estimate": 0.004023242451542919
     },
     {
       "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": 29025.999115964976,
-      "required_pages_per_second_upper_estimate": 0.0017400643268676694
+      "required_token_positions_per_second": 28975.649623529,
+      "required_pages_per_second_upper_estimate": 0.001737045951675356
     },
     {
       "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": 29025.999115964976,
-      "required_pages_per_second_upper_estimate": 0.002256643759388946
+      "required_token_positions_per_second": 28975.649623529,
+      "required_pages_per_second_upper_estimate": 0.0022527293078160606
     },
     {
       "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": 29025.999115964976,
-      "required_pages_per_second_upper_estimate": 0.3723143509699077
+      "required_token_positions_per_second": 28975.649623529,
+      "required_pages_per_second_upper_estimate": 0.37166852174201204
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12273238110185503,
-    "producer_p95_seconds": 0.268817897002009,
-    "gpu_update_seconds": 0.9525253511357324,
-    "producer_batches_per_second": 8.147809005433576,
-    "gpu_batches_per_second": 1.0498408245068351,
-    "producer_headroom": 7.7609946338875,
+    "producer_batch_seconds": 0.12662399446097083,
+    "producer_p95_seconds": 0.27011396599846194,
+    "gpu_update_seconds": 0.9541805053284841,
+    "producer_batches_per_second": 7.897397363405944,
+    "gpu_batches_per_second": 1.0480197346473161,
+    "producer_headroom": 7.535542606994522,
     "sustainable": true,
-    "buffer_seconds": 60.00909712155114,
+    "buffer_seconds": 60.113371835694494,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 459567.3219617912,
+    "generation_bytes_per_second": 470550.2745875754,
     "upload_bytes_per_second": 5601306.440894706,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.6124491132765,
+    "estimated_full_shards_per_hour": 100.96913507671782,
     "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-v1/1790914434605211575-090606d4/train.jsonl

@@ -218,3 +218,4 @@
 {"step": 506800, "loss": 0.5715804100036621, "lr": 0.0003, "grad_norm": 0.1542340269721092, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74178678963, "tokens_per_second": 150929.81368149302, "data_wait_s": 9.966999641619623e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 506900, "loss": 0.5570117235183716, "lr": 0.0003, "grad_norm": 0.15393902032819415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74193322151, "tokens_per_second": 152792.72910903775, "data_wait_s": 2.1479998395079747e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 507000, "loss": 0.5756789445877075, "lr": 0.0003, "grad_norm": 0.165578137187598, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74207972093, "tokens_per_second": 152320.78373612696, "data_wait_s": 1.134999911300838e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 507100, "loss": 0.6167461276054382, "lr": 0.0003, "grad_norm": 0.17118224313750252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74222571305, "tokens_per_second": 150993.34753739194, "data_wait_s": 1.2050000805174932e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2529905f8518ea531323dcb8f8a95491e4076b2579d12c17127dd4e1d649935f
-size 469257
+oid sha256:99b92c54b5f957f06ad297a38efbe5b825f33a8c6abad16c89243c3faabd63a0
+size 471375