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": 77369.052332473,
-      "required_pages_per_second_upper_estimate": 0.005352570561295051
+      "required_token_positions_per_second": 77312.46282452886,
+      "required_pages_per_second_upper_estimate": 0.005348655567829728
     },
     {
       "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": 9671.131541559125,
-      "required_pages_per_second_upper_estimate": 0.004049100820259014
+      "required_token_positions_per_second": 9664.057853066108,
+      "required_pages_per_second_upper_estimate": 0.004046139214602401
     },
     {
       "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": 19342.26308311825,
-      "required_pages_per_second_upper_estimate": 0.004028483310433977
+      "required_token_positions_per_second": 19328.115706132216,
+      "required_pages_per_second_upper_estimate": 0.004025536784899207
     },
     {
       "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": 29013.394624677378,
-      "required_pages_per_second_upper_estimate": 0.0017393087068609215
+      "required_token_positions_per_second": 28992.173559198323,
+      "required_pages_per_second_upper_estimate": 0.001738036536388143
     },
     {
       "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": 29013.394624677378,
-      "required_pages_per_second_upper_estimate": 0.002255663815632633
+      "required_token_positions_per_second": 28992.173559198323,
+      "required_pages_per_second_upper_estimate": 0.002254013971133239
     },
     {
       "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": 29013.394624677378,
-      "required_pages_per_second_upper_estimate": 0.37215267408931874
+      "required_token_positions_per_second": 28992.173559198323,
+      "required_pages_per_second_upper_estimate": 0.3718804730467583
     }
   ],
   "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": 0.08060388910178062,
-    "producer_p95_seconds": 0.22325702699890826,
-    "gpu_update_seconds": 0.9529391633643572,
-    "producer_batches_per_second": 12.40634926110419,
-    "gpu_batches_per_second": 1.0493849328948703,
-    "producer_headroom": 11.822496085282637,
+    "producer_batch_seconds": 0.06540388839081857,
+    "producer_p95_seconds": 0.2294216049995157,
+    "gpu_update_seconds": 0.9536366752063727,
+    "producer_batches_per_second": 15.289610825957872,
+    "gpu_batches_per_second": 1.0486173885705412,
+    "producer_headroom": 14.58073363326583,
     "sustainable": true,
-    "buffer_seconds": 60.03516729195451,
+    "buffer_seconds": 60.07911053800148,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384295.560956119,
-    "upload_bytes_per_second": 5552634.537313602,
+    "generation_bytes_per_second": 385672.25190008257,
+    "upload_bytes_per_second": 5547138.020430888,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.4608814383762,
+    "estimated_full_shards_per_hour": 82.75628726722582,
     "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": 0.9979469910012995,
-      "bytes": 6506963,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1795985310000106,
       "bytes": 8204299,
@@ -562,6 +556,12 @@
       "bytes": 2636885,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.006605718001083,
+      "bytes": 6127707,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -174,3 +174,4 @@
 {"step": 502400, "loss": 0.6325347423553467, "lr": 0.0003, "grad_norm": 0.20552976388713154, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73534579820, "tokens_per_second": 153798.08455313052, "data_wait_s": 9.646999387769029e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 502500, "loss": 0.5735471248626709, "lr": 0.0003, "grad_norm": 0.15704754788065672, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73549106769, "tokens_per_second": 150982.9314653788, "data_wait_s": 1.1591000657062978e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 502600, "loss": 0.5724793076515198, "lr": 0.0003, "grad_norm": 0.1434232357324958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73563901469, "tokens_per_second": 154233.70855758447, "data_wait_s": 1.748100112308748e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 502700, "loss": 0.537597119808197, "lr": 0.0003, "grad_norm": 0.14868456586983175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73578574033, "tokens_per_second": 152649.4251514213, "data_wait_s": 1.2917000276502222e-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 sha256:3b44465dc2bf3f0c0a39d240d3bf42c7e22a6472ec626f052f78a2262d16d3d7
-size 375282
+oid sha256:8d41a831ed7d1eb87083a818ff88004e20a9aea29b509ce5472a1e880b35b251
+size 377400