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": 75754.55234790196,
-      "required_pages_per_second_upper_estimate": 0.006566212599358549
+      "required_token_positions_per_second": 75531.15246238127,
+      "required_pages_per_second_upper_estimate": 0.006546848863483449
     },
     {
       "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": 9469.319043487745,
-      "required_pages_per_second_upper_estimate": 0.0037774529453836543
+      "required_token_positions_per_second": 9441.39405779766,
+      "required_pages_per_second_upper_estimate": 0.0037663132510761367
     },
     {
       "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": 18938.63808697549,
-      "required_pages_per_second_upper_estimate": 0.004381120612775353
+      "required_token_positions_per_second": 18882.78811559532,
+      "required_pages_per_second_upper_estimate": 0.004368200704822475
     },
     {
       "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": 28407.957130463234,
-      "required_pages_per_second_upper_estimate": 0.0017298955442129544
+      "required_token_positions_per_second": 28324.18217339298,
+      "required_pages_per_second_upper_estimate": 0.0017247940888606068
     },
     {
       "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": 28407.957130463234,
-      "required_pages_per_second_upper_estimate": 0.0021869231781614694
+      "required_token_positions_per_second": 28324.18217339298,
+      "required_pages_per_second_upper_estimate": 0.0021804739500622776
     },
     {
       "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": 28407.957130463234,
-      "required_pages_per_second_upper_estimate": 0.35273175224385356
+      "required_token_positions_per_second": 28324.18217339298,
+      "required_pages_per_second_upper_estimate": 0.3516915476537862
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09591713410975444,
-    "producer_p95_seconds": 0.3745843260039692,
-    "gpu_update_seconds": 0.9732484413795354,
-    "producer_batches_per_second": 10.425665959282487,
-    "gpu_batches_per_second": 1.0274868753784445,
-    "producer_headroom": 10.14676314521536,
+    "producer_batch_seconds": 0.15292593792952403,
+    "producer_p95_seconds": 0.4020387929995195,
+    "gpu_update_seconds": 0.9761270362810981,
+    "producer_batches_per_second": 6.539113073551005,
+    "gpu_batches_per_second": 1.0244568205075586,
+    "producer_headroom": 6.383005064392324,
     "sustainable": true,
-    "buffer_seconds": 60.34140336553119,
+    "buffer_seconds": 60.51987624942808,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366766.4687634059,
-    "upload_bytes_per_second": 4234795.073533338,
+    "generation_bytes_per_second": 367075.5050695447,
+    "upload_bytes_per_second": 4242088.359833767,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.69954631020197,
+    "estimated_full_shards_per_hour": 78.76585830750233,
     "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": 0.9470581830028095,
-      "bytes": 5899775,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.0518160799983889,
       "bytes": 1140036,
@@ -562,6 +556,12 @@
       "bytes": 6059900,
       "pages": 26,
       "shards": 2
+    },
+    {
+      "seconds": 1.0482589550010744,
+      "bytes": 7056720,
+      "pages": 42,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -344,3 +344,4 @@
 {"step": 34400, "loss": 0.6432075276970863, "lr": 0.0003, "grad_norm": 0.13552194736432954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5033696260, "tokens_per_second": 146893.72778329093, "data_wait_s": 1.0436997399665415e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34500, "loss": 0.6013185773044825, "lr": 0.0003, "grad_norm": 0.13190112491699676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5048308953, "tokens_per_second": 148703.76954503203, "data_wait_s": 1.3197000953368843e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34600, "loss": 0.6039976011961699, "lr": 0.0003, "grad_norm": 0.1353956471650447, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5063136574, "tokens_per_second": 151591.29069847715, "data_wait_s": 1.2245996913406998e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 34700, "loss": 0.672161877155304, "lr": 0.0003, "grad_norm": 0.12548074364156325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5077686867, "tokens_per_second": 147787.68587057784, "data_wait_s": 1.098900247598067e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8d650a97e8ac6e79d7d85e98d39cbf3db2112d67fd6d9f6153c878fcb09223f4
-size 731917
+oid sha256:398182d1c579d6d76147869898353319f7db44254ceba899a52554e3d0e63b82
+size 734035