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": 74933.80795274138,
-      "required_pages_per_second_upper_estimate": 0.006495072555343681
+      "required_token_positions_per_second": 74281.70716910435,
+      "required_pages_per_second_upper_estimate": 0.00643855011215235
     },
     {
       "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": 9366.725994092672,
-      "required_pages_per_second_upper_estimate": 0.003736527044077179
+      "required_token_positions_per_second": 9285.213396138044,
+      "required_pages_per_second_upper_estimate": 0.003704010450031133
     },
     {
       "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": 18733.451988185345,
-      "required_pages_per_second_upper_estimate": 0.004333654419972255
+      "required_token_positions_per_second": 18570.42679227609,
+      "required_pages_per_second_upper_estimate": 0.004295941410044103
     },
     {
       "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": 28100.17798227802,
-      "required_pages_per_second_upper_estimate": 0.0017111534088808677
+      "required_token_positions_per_second": 27855.640188414134,
+      "required_pages_per_second_upper_estimate": 0.0016962623402252068
     },
     {
       "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": 28100.17798227802,
-      "required_pages_per_second_upper_estimate": 0.0021632294873469597
+      "required_token_positions_per_second": 27855.640188414134,
+      "required_pages_per_second_upper_estimate": 0.0021444042910513795
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28100.17798227802,
-      "required_pages_per_second_upper_estimate": 0.34891016529393964
+      "required_token_positions_per_second": 27855.640188414134,
+      "required_pages_per_second_upper_estimate": 0.34587382430949914
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.07160476486718181,
-    "producer_p95_seconds": 0.18144677499913087,
-    "gpu_update_seconds": 0.9839083587811012,
-    "producer_batches_per_second": 13.965551061509373,
-    "gpu_batches_per_second": 1.0163548170673473,
-    "producer_headroom": 13.740822424403351,
+    "producer_batch_seconds": 0.07732148725784782,
+    "producer_p95_seconds": 0.209619295000266,
+    "gpu_update_seconds": 0.9925458475551211,
+    "producer_batches_per_second": 12.933015588089377,
+    "gpu_batches_per_second": 1.0075101341295623,
+    "producer_headroom": 12.836610918323764,
     "sustainable": true,
-    "buffer_seconds": 60.01840988564717,
+    "buffer_seconds": 60.54529670086239,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 400500.12039787456,
-    "upload_bytes_per_second": 5098133.898308832,
+    "generation_bytes_per_second": 384601.5162067605,
+    "upload_bytes_per_second": 5149305.390704784,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.93800267173937,
+    "estimated_full_shards_per_hour": 82.52653231289015,
     "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": 7523830,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 0.8949287559998993,
+      "bytes": 6264749,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 161000, "loss": 0.5677636433392763, "lr": 0.0003, "grad_norm": 0.14378056862540295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23563526151, "tokens_per_second": 148497.77133646054, "data_wait_s": 1.337000048806658e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 161100, "loss": 0.5860947147011757, "lr": 0.0003, "grad_norm": 0.1354543352136411, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23577993896, "tokens_per_second": 147977.07669252725, "data_wait_s": 1.991199951589806e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 161200, "loss": 0.5865148399025202, "lr": 0.0003, "grad_norm": 0.13777146085472525, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23592569679, "tokens_per_second": 147987.39522450994, "data_wait_s": 1.0426000699226279e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 161300, "loss": 0.6452075373381376, "lr": 0.0003, "grad_norm": 0.14168178188903088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 23607329339, "tokens_per_second": 151364.45356469764, "data_wait_s": 1.2057000276399776e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5d2fb285a0d43ad79bd3e7ea21bd5c9ec6b1c757777a058f7c14958dbf82fc26
-size 132264
+oid sha256:a7e32846af6f65d2482d1183a8f11462f2bc1f2f29454482dff9d13b7d77eeed
+size 134382