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": 76669.36261268068,
-      "required_pages_per_second_upper_estimate": 0.005304164402975994
+      "required_token_positions_per_second": 76688.27427350907,
+      "required_pages_per_second_upper_estimate": 0.005305472755553201
     },
     {
       "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": 9583.670326585085,
-      "required_pages_per_second_upper_estimate": 0.004012482635947238
+      "required_token_positions_per_second": 9586.034284188634,
+      "required_pages_per_second_upper_estimate": 0.004013472375630796
     },
     {
       "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": 19167.34065317017,
-      "required_pages_per_second_upper_estimate": 0.003992051581290482
+      "required_token_positions_per_second": 19172.068568377268,
+      "required_pages_per_second_upper_estimate": 0.003993036281344612
     },
     {
       "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": 28751.010979755254,
-      "required_pages_per_second_upper_estimate": 0.0017235792079844684
+      "required_token_positions_per_second": 28758.102852565902,
+      "required_pages_per_second_upper_estimate": 0.001724004354930804
     },
     {
       "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": 28751.010979755254,
-      "required_pages_per_second_upper_estimate": 0.002235264641343622
+      "required_token_positions_per_second": 28758.102852565902,
+      "required_pages_per_second_upper_estimate": 0.0022358160032607986
     },
     {
       "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": 28751.010979755254,
-      "required_pages_per_second_upper_estimate": 0.36878709841786605
+      "required_token_positions_per_second": 28758.102852565902,
+      "required_pages_per_second_upper_estimate": 0.3688780653476213
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08056415094517888,
-    "producer_p95_seconds": 0.2960309679983766,
-    "gpu_update_seconds": 0.9616357497643498,
-    "producer_batches_per_second": 12.412468675806755,
-    "gpu_batches_per_second": 1.0398947837006385,
-    "producer_headroom": 11.936273621485935,
+    "producer_batch_seconds": 0.4087949200703065,
+    "producer_p95_seconds": 0.3108744620003563,
+    "gpu_update_seconds": 0.9613986062204081,
+    "producer_batches_per_second": 2.4462143507752376,
+    "gpu_batches_per_second": 1.040151289516996,
+    "producer_headroom": 2.351787067351674,
     "sustainable": true,
-    "buffer_seconds": 60.58305223515404,
+    "buffer_seconds": 60.56811219188571,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268698.82392199844,
+    "generation_bytes_per_second": 256373.05799625575,
     "upload_bytes_per_second": 4539492.443985672,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.65651262516942,
+    "estimated_full_shards_per_hour": 55.011690186650796,
     "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-v1/1791342976705245012-6808dd50/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 872000, "loss": 0.5325416922569275, "lr": 3e-05, "grad_norm": 0.2988365468834813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127636444537, "tokens_per_second": 154200.01939884928, "data_wait_s": 1.575900023453869e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872100, "loss": 0.5009547472000122, "lr": 3e-05, "grad_norm": 0.30272764242403954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127651203827, "tokens_per_second": 132028.77519436993, "data_wait_s": 1.2270000297576189e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 872200, "loss": 0.5043037533760071, "lr": 3e-05, "grad_norm": 0.30055156094455737, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127665666258, "tokens_per_second": 149899.29615613303, "data_wait_s": 1.6519999917363748e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 872300, "loss": 0.4845273494720459, "lr": 3e-05, "grad_norm": 0.31965501601472557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127680161674, "tokens_per_second": 150525.67748754594, "data_wait_s": 1.1422998795751482e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a20f76ac03f0861c45df4d69b4bc3e84a5d59567959a657c12ae2148f2c1327
-size 633909
+oid sha256:e117f5d3cd910d438fe76dd12c4d5e2c7c7e62882c2e1f6d1f7c57983974b847
+size 636027