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": 76536.61347537351,
-      "required_pages_per_second_upper_estimate": 0.005294980509636648
+      "required_token_positions_per_second": 77402.66430923995,
+      "required_pages_per_second_upper_estimate": 0.00535489591583832
     },
     {
       "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": 9567.076684421689,
-      "required_pages_per_second_upper_estimate": 0.004005535224488077
+      "required_token_positions_per_second": 9675.333038654993,
+      "required_pages_per_second_upper_estimate": 0.004050859899355818
     },
     {
       "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": 19134.153368843377,
-      "required_pages_per_second_upper_estimate": 0.003985139545172754
+      "required_token_positions_per_second": 19350.666077309987,
+      "required_pages_per_second_upper_estimate": 0.004030233432522258
     },
     {
       "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": 28701.23005326507,
-      "required_pages_per_second_upper_estimate": 0.0017205949174524554
+      "required_token_positions_per_second": 29025.999115964976,
+      "required_pages_per_second_upper_estimate": 0.0017400643268676694
     },
     {
       "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": 28701.23005326507,
-      "required_pages_per_second_upper_estimate": 0.0022313943932721655
+      "required_token_positions_per_second": 29025.999115964976,
+      "required_pages_per_second_upper_estimate": 0.002256643759388946
     },
     {
       "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": 28701.23005326507,
-      "required_pages_per_second_upper_estimate": 0.36814856214344444
+      "required_token_positions_per_second": 29025.999115964976,
+      "required_pages_per_second_upper_estimate": 0.3723143509699077
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07051779232031663,
-    "producer_p95_seconds": 0.2533027809986379,
-    "gpu_update_seconds": 0.9633036615047357,
-    "producer_batches_per_second": 14.180818302672439,
-    "gpu_batches_per_second": 1.038094258292284,
-    "producer_headroom": 13.66043419409773,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.688130674798344,
+    "buffer_seconds": 60.00909712155114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 444504.35773233767,
-    "upload_bytes_per_second": 5569695.701669307,
+    "generation_bytes_per_second": 459567.3219617912,
+    "upload_bytes_per_second": 5601306.440894706,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.38028763749692,
+    "estimated_full_shards_per_hour": 98.6124491132765,
     "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": 0.8494176380008867,
-      "bytes": 947075,
-      "pages": 27,
-      "shards": 1
-    },
     {
       "seconds": 1.246135413999582,
       "bytes": 6441176,
@@ -562,6 +556,12 @@
       "bytes": 11895475,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 1.5734916069995961,
+      "bytes": 7449351,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -217,3 +217,4 @@
 {"step": 506700, "loss": 0.5619561076164246, "lr": 0.0003, "grad_norm": 0.1554463108727053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 74164155673, "tokens_per_second": 152544.85657892728, "data_wait_s": 1.4468998415395617e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:0c3b61aaa3a00d1afdf78b8576a84cc6a5685f33f3fa731a2202af0e89f242aa
-size 467139
+oid sha256:2529905f8518ea531323dcb8f8a95491e4076b2579d12c17127dd4e1d649935f
+size 469257