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": 76382.5577012021,
-      "required_pages_per_second_upper_estimate": 0.0066206465117175884
+      "required_token_positions_per_second": 76296.12447567155,
+      "required_pages_per_second_upper_estimate": 0.0066131546987916075
     },
     {
       "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": 9547.819712650262,
-      "required_pages_per_second_upper_estimate": 0.003808768036002179
+      "required_token_positions_per_second": 9537.015559458943,
+      "required_pages_per_second_upper_estimate": 0.0038044580977576762
     },
     {
       "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": 19095.639425300524,
-      "required_pages_per_second_upper_estimate": 0.004417440109267667
+      "required_token_positions_per_second": 19074.031118917886,
+      "required_pages_per_second_upper_estimate": 0.004412441407879248
     },
     {
       "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": 28643.459137950784,
-      "required_pages_per_second_upper_estimate": 0.0017442363808854039
+      "required_token_positions_per_second": 28611.046678376828,
+      "required_pages_per_second_upper_estimate": 0.0017422626321523828
     },
     {
       "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": 28643.459137950784,
-      "required_pages_per_second_upper_estimate": 0.002205052774609142
+      "required_token_positions_per_second": 28611.046678376828,
+      "required_pages_per_second_upper_estimate": 0.002202557573747708
     },
     {
       "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": 28643.459137950784,
-      "required_pages_per_second_upper_estimate": 0.35565589900233163
+      "required_token_positions_per_second": 28611.046678376828,
+      "required_pages_per_second_upper_estimate": 0.35525344473194714
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.05815594388270995,
-    "producer_p95_seconds": 0.19432858800064423,
-    "gpu_update_seconds": 0.9652465460558894,
-    "producer_batches_per_second": 17.195146931443837,
-    "gpu_batches_per_second": 1.0360047431261135,
-    "producer_headroom": 16.59755618449969,
+    "producer_batch_seconds": 0.07955688428899066,
+    "producer_p95_seconds": 0.1956174259976251,
+    "gpu_update_seconds": 0.9663400402927355,
+    "producer_batches_per_second": 12.569622464945919,
+    "gpu_batches_per_second": 1.0348324174760137,
+    "producer_headroom": 12.146529479240312,
     "sustainable": true,
-    "buffer_seconds": 60.81053240152104,
+    "buffer_seconds": 60.879422538442334,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 420706.99569984904,
-    "upload_bytes_per_second": 5735663.990811707,
+    "generation_bytes_per_second": 415473.4372649945,
+    "upload_bytes_per_second": 5664772.007316267,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.27392771956066,
+    "estimated_full_shards_per_hour": 89.15092791044594,
     "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": 1.407603981999273,
-      "bytes": 11506972,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.4075674270006857,
       "bytes": 6094729,
@@ -562,6 +556,12 @@
       "bytes": 12406942,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.4563389440008905,
+      "bytes": 6496961,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 288400, "loss": 0.571783060207963, "lr": 0.0003, "grad_norm": 0.14598619365887963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42213362594, "tokens_per_second": 148148.08819389716, "data_wait_s": 1.1122998330392875e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 288500, "loss": 0.6058273836970329, "lr": 0.0003, "grad_norm": 0.1596113652324245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42227982364, "tokens_per_second": 150208.51362881463, "data_wait_s": 1.1613003152888268e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 288600, "loss": 0.5930606629699469, "lr": 0.0003, "grad_norm": 0.14580331171454478, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42242601227, "tokens_per_second": 150559.58680798727, "data_wait_s": 1.3025000953348354e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 288700, "loss": 0.5737161673605442, "lr": 0.0003, "grad_norm": 0.1530994809562534, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 42257440090, "tokens_per_second": 152680.61761010755, "data_wait_s": 1.4662000467069447e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ce4058fc58187b40bd13547afa5b647920ae1bb393c83cb88d48a6b3e96d1c1
-size 178860
+oid sha256:714db44a1680cd7dfa978c6931af96db17ea5c864c8ff1e36467f3f9c893c261
+size 180978