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": 75841.6080887182,
-      "required_pages_per_second_upper_estimate": 0.0065737583703317426
+      "required_token_positions_per_second": 74994.94563402461,
+      "required_pages_per_second_upper_estimate": 0.006500371814605275
     },
     {
       "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": 9480.201011089775,
-      "required_pages_per_second_upper_estimate": 0.003781793924960019
+      "required_token_positions_per_second": 9374.368204253076,
+      "required_pages_per_second_upper_estimate": 0.003739575635971388
     },
     {
       "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": 18960.40202217955,
-      "required_pages_per_second_upper_estimate": 0.0043861553161526345
+      "required_token_positions_per_second": 18748.736408506153,
+      "required_pages_per_second_upper_estimate": 0.004337190201616864
     },
     {
       "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": 28440.603033269326,
-      "required_pages_per_second_upper_estimate": 0.0017318835084140305
+      "required_token_positions_per_second": 28123.10461275923,
+      "required_pages_per_second_upper_estimate": 0.0017125495203904396
     },
     {
       "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": 28440.603033269326,
-      "required_pages_per_second_upper_estimate": 0.0021894363501291252
+      "required_token_positions_per_second": 28123.10461275923,
+      "required_pages_per_second_upper_estimate": 0.0021649944428263784
     },
     {
       "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": 28440.603033269326,
-      "required_pages_per_second_upper_estimate": 0.3531371050978969
+      "required_token_positions_per_second": 28123.10461275923,
+      "required_pages_per_second_upper_estimate": 0.34919483731402
     }
   ],
   "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": 0.07889537953909098,
-    "producer_p95_seconds": 0.2192219059998024,
-    "gpu_update_seconds": 0.9721312859526167,
-    "producer_batches_per_second": 12.675013490549485,
-    "gpu_batches_per_second": 1.0286676444324843,
-    "producer_headroom": 12.321777164034636,
+    "producer_batch_seconds": 0.43148350484376863,
+    "producer_p95_seconds": 0.23092953500054136,
+    "gpu_update_seconds": 0.9831062530506081,
+    "producer_batches_per_second": 2.3175856985821035,
+    "gpu_batches_per_second": 1.017184049940655,
+    "producer_headroom": 2.278432992256728,
     "sustainable": true,
-    "buffer_seconds": 60.27213972906223,
+    "buffer_seconds": 60.9525876891377,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347442.16707860184,
-    "upload_bytes_per_second": 3965630.717384473,
+    "generation_bytes_per_second": 309592.109125265,
+    "upload_bytes_per_second": 3981097.7090012296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.55300101536314,
+    "estimated_full_shards_per_hour": 66.4312596828314,
     "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": 1.2328179739997722,
-      "bytes": 5713232,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.9817969460000313,
       "bytes": 2633705,
@@ -562,6 +556,12 @@
       "bytes": 2416731,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.9309691420003219,
+      "bytes": 6123756,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -118,3 +118,4 @@
 {"step": 326800, "loss": 0.5700195319950581, "lr": 0.0003, "grad_norm": 0.14546542419295197, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47834562939, "tokens_per_second": 147930.21163798554, "data_wait_s": 1.1394000466680154e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 326900, "loss": 0.6353239305317402, "lr": 0.0003, "grad_norm": 0.18132119997884796, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47849120395, "tokens_per_second": 148838.9649229429, "data_wait_s": 1.4277000445872545e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 327000, "loss": 0.5851198229938745, "lr": 0.0003, "grad_norm": 0.17447375703832715, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47863871304, "tokens_per_second": 149746.83341947396, "data_wait_s": 1.3969000065117143e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 327100, "loss": 0.5740393213927746, "lr": 0.0003, "grad_norm": 0.14900409334726034, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 47878431430, "tokens_per_second": 148157.81349554347, "data_wait_s": 1.923099989653565e-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:d9f6f6b13e8cd439bc1b3539c4b78304b9b19005074c6d148d5893c4717b75bb
-size 255891
+oid sha256:a7fbf58943696ed31a21b612203d30ea13373c7bf99c9fcd09590b2d30dd877e
+size 258009