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": 75087.94175655725,
-      "required_pages_per_second_upper_estimate": 0.0065084324828098864
+      "required_token_positions_per_second": 75631.06268419111,
+      "required_pages_per_second_upper_estimate": 0.006555508828289784
     },
     {
       "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": 9385.992719569656,
-      "required_pages_per_second_upper_estimate": 0.0037442128289331645
+      "required_token_positions_per_second": 9453.88283552389,
+      "required_pages_per_second_upper_estimate": 0.0037712952112349965
     },
     {
       "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": 18771.985439139313,
-      "required_pages_per_second_upper_estimate": 0.004342568455684881
+      "required_token_positions_per_second": 18907.76567104778,
+      "required_pages_per_second_upper_estimate": 0.0043739788226864395
     },
     {
       "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": 28157.978158708967,
-      "required_pages_per_second_upper_estimate": 0.0017146731363714234
+      "required_token_positions_per_second": 28361.64850657167,
+      "required_pages_per_second_upper_estimate": 0.0017270755919805306
     },
     {
       "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": 28157.978158708967,
-      "required_pages_per_second_upper_estimate": 0.002167679104929743
+      "required_token_positions_per_second": 28361.64850657167,
+      "required_pages_per_second_upper_estimate": 0.0021833582121038213
     },
     {
       "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": 28157.978158708967,
-      "required_pages_per_second_upper_estimate": 0.3496278500404655
+      "required_token_positions_per_second": 28361.64850657167,
+      "required_pages_per_second_upper_estimate": 0.3521567541201146
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.126282340062005,
-    "producer_p95_seconds": 0.3703979839992826,
-    "gpu_update_seconds": 0.9818886797967333,
-    "producer_batches_per_second": 7.918763617375138,
-    "gpu_batches_per_second": 1.0184453905783046,
-    "producer_headroom": 7.7753443538868785,
+    "producer_batch_seconds": 0.10986396203907134,
+    "producer_p95_seconds": 0.3852085789985722,
+    "gpu_update_seconds": 0.9748375519707075,
+    "producer_batches_per_second": 9.10216581889124,
+    "gpu_batches_per_second": 1.0258119396184777,
+    "producer_headroom": 8.873133044519388,
     "sustainable": true,
-    "buffer_seconds": 60.87709814739747,
+    "buffer_seconds": 60.439928222183866,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319275.4862760083,
-    "upload_bytes_per_second": 3275875.5693956525,
+    "generation_bytes_per_second": 353916.43879189505,
+    "upload_bytes_per_second": 3167593.213277648,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.50908700189768,
+    "estimated_full_shards_per_hour": 75.94222901170386,
     "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": 1.7832624309994571,
-      "bytes": 17985234,
-      "pages": 39,
-      "shards": 5
-    },
     {
       "seconds": 0.969149242999265,
       "bytes": 1068713,
@@ -562,6 +556,12 @@
       "bytes": 7986291,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.8759991519982577,
+      "bytes": 805335,
+      "pages": 26,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 347800, "loss": 0.5847104731947184, "lr": 0.0003, "grad_norm": 0.1490548749841576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50909009148, "tokens_per_second": 149800.97768517808, "data_wait_s": 1.2564996723085642e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 347900, "loss": 0.5391279589384794, "lr": 0.0003, "grad_norm": 0.13975359883808136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50923617558, "tokens_per_second": 148858.53260996024, "data_wait_s": 1.3570002920459956e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 348000, "loss": 0.5806275233626366, "lr": 0.0003, "grad_norm": 0.1628066548025042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50938233409, "tokens_per_second": 149651.37517678246, "data_wait_s": 1.3744000170845538e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 348100, "loss": 0.5532508082687855, "lr": 0.0003, "grad_norm": 0.1465484121571448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 50952768903, "tokens_per_second": 145933.23827819858, "data_wait_s": 1.0547999409027398e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9eb96d3541fbb4f65053d3e7ade7be548cae29fc79a4bf7207c1f714b1a7542c
-size 703803
+oid sha256:196c13d80b840b6c0fd8867f03049eea9c43ba5a5a7de5fc3b796ed678b464d4
+size 705921