Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75635.51206845751,
-      "required_pages_per_second_upper_estimate": 0.006555894489641134
+      "required_token_positions_per_second": 75770.1707126706,
+      "required_pages_per_second_upper_estimate": 0.006567566359632311
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9454.439008557189,
-      "required_pages_per_second_upper_estimate": 0.0037715170769735077
+      "required_token_positions_per_second": 9471.271339083825,
+      "required_pages_per_second_upper_estimate": 0.0037782317452863512
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18908.878017114377,
-      "required_pages_per_second_upper_estimate": 0.004374236144372325
+      "required_token_positions_per_second": 18942.54267816765,
+      "required_pages_per_second_upper_estimate": 0.004382023871229201
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28363.317025671564,
-      "required_pages_per_second_upper_estimate": 0.001727177196039671
+      "required_token_positions_per_second": 28413.814017251476,
+      "required_pages_per_second_upper_estimate": 0.0017302521978896483
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28363.317025671564,
-      "required_pages_per_second_upper_estimate": 0.0021834866593228876
+      "required_token_positions_per_second": 28413.814017251476,
+      "required_pages_per_second_upper_estimate": 0.0021873740575193285
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28363.317025671564,
-      "required_pages_per_second_upper_estimate": 0.3521774715431611
+      "required_token_positions_per_second": 28413.814017251476,
+      "required_pages_per_second_upper_estimate": 0.35280447517602437
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2583886677812828,
-    "producer_p95_seconds": 0.1763358389998757,
-    "gpu_update_seconds": 0.9747802055371686,
-    "producer_batches_per_second": 3.8701387664820737,
-    "gpu_batches_per_second": 1.0258722882549032,
-    "producer_headroom": 3.7725346622487597,
+    "producer_batch_seconds": 0.410958472664106,
+    "producer_p95_seconds": 0.17358576500009804,
+    "gpu_update_seconds": 0.9730478274832618,
+    "producer_batches_per_second": 2.4333358879726585,
+    "gpu_batches_per_second": 1.0276987130082276,
+    "producer_headroom": 2.3677521993288493,
     "sustainable": true,
-    "buffer_seconds": 60.43637274330445,
+    "buffer_seconds": 60.32896530396223,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390983.68114543805,
-    "upload_bytes_per_second": 6021083.6783039775,
+    "generation_bytes_per_second": 370384.72083584504,
+    "upload_bytes_per_second": 6001769.56377775,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.89599633953434,
+    "estimated_full_shards_per_hour": 79.47593897634997,
     "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": [
@@ -334,6 +334,12 @@
       "bytes": 5940010,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 0.9267724249993989,
+      "bytes": 5041964,
+      "pages": 22,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 45600, "loss": 0.6086126398295164, "lr": 0.0003, "grad_norm": 0.11961747347757953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6672697519, "tokens_per_second": 149299.29169207055, "data_wait_s": 1.3101999684295151e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 45700, "loss": 0.6558852419257164, "lr": 0.0003, "grad_norm": 0.1374799079946057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6687336151, "tokens_per_second": 148993.07911957163, "data_wait_s": 9.60300076258136e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 45800, "loss": 0.6263294331729412, "lr": 0.0003, "grad_norm": 0.12379982046226583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6702066400, "tokens_per_second": 150151.61350128875, "data_wait_s": 1.4790000022912864e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 45900, "loss": 0.6565524749457836, "lr": 0.0003, "grad_norm": 0.13112385998803777, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6716776605, "tokens_per_second": 150495.40776296184, "data_wait_s": 1.0531999578233808e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d9fb1052d6dd0c4aff66153fdd6ccc6bfcb3ecb4f9adf4be3ac5e5d452c9efed
-size 123792
+oid sha256:5dfca27a6dfbd3aa76acab2685e584062dadbe710d5ddbb96f355e65930b5dff
+size 125910