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": 75735.32456221439,
-      "required_pages_per_second_upper_estimate": 0.006564545983627527
+      "required_token_positions_per_second": 75781.1675313154,
+      "required_pages_per_second_upper_estimate": 0.00656851953600126
     },
     {
       "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": 9466.915570276798,
-      "required_pages_per_second_upper_estimate": 0.003776494163984681
+      "required_token_positions_per_second": 9472.645941414425,
+      "required_pages_per_second_upper_estimate": 0.003778780094708164
     },
     {
       "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": 18933.831140553597,
-      "required_pages_per_second_upper_estimate": 0.004380008610320015
+      "required_token_positions_per_second": 18945.29188282885,
+      "required_pages_per_second_upper_estimate": 0.0043826598513432196
     },
     {
       "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": 28400.746710830397,
-      "required_pages_per_second_upper_estimate": 0.001729456467487463
+      "required_token_positions_per_second": 28417.937824243276,
+      "required_pages_per_second_upper_estimate": 0.0017305033160995057
     },
     {
       "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": 28400.746710830397,
-      "required_pages_per_second_upper_estimate": 0.0021863681000984145
+      "required_token_positions_per_second": 28417.937824243276,
+      "required_pages_per_second_upper_estimate": 0.0021876915195969855
     },
     {
       "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": 28400.746710830397,
-      "required_pages_per_second_upper_estimate": 0.3526422229637359
+      "required_token_positions_per_second": 28417.937824243276,
+      "required_pages_per_second_upper_estimate": 0.35285567905736837
     }
   ],
   "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": 0.1140211412579788,
-    "producer_p95_seconds": 0.38258683999447385,
-    "gpu_update_seconds": 0.9734955309980162,
-    "producer_batches_per_second": 8.770303375033299,
-    "gpu_batches_per_second": 1.0272260818442707,
-    "producer_headroom": 8.537851141091735,
+    "producer_batch_seconds": 0.09915572425057917,
+    "producer_p95_seconds": 0.3918896820032387,
+    "gpu_update_seconds": 0.9729066257725977,
+    "producer_batches_per_second": 10.085146445734917,
+    "gpu_batches_per_second": 1.027847866906947,
+    "producer_headroom": 9.811905798942464,
     "sustainable": true,
-    "buffer_seconds": 60.35672292187701,
+    "buffer_seconds": 60.320210797901055,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291151.9316224632,
-    "upload_bytes_per_second": 5697490.247156283,
+    "generation_bytes_per_second": 290164.35848629964,
+    "upload_bytes_per_second": 5703744.979600916,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.47442685609267,
+    "estimated_full_shards_per_hour": 62.26251665059797,
     "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": 0.8875501189977513,
-      "bytes": 1102979,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.025324021000415,
       "bytes": 6368505,
@@ -562,6 +556,12 @@
       "bytes": 12510715,
       "pages": 47,
       "shards": 4
+    },
+    {
+      "seconds": 0.7931236949953018,
+      "bytes": 1028533,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 234600, "loss": 0.5857234466820955, "lr": 0.0003, "grad_norm": 0.1330573131601293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34338929386, "tokens_per_second": 149179.78533397504, "data_wait_s": 8.201997843571007e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 234700, "loss": 0.5773926191031933, "lr": 0.0003, "grad_norm": 0.1411813916132267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34353590868, "tokens_per_second": 149094.09296764847, "data_wait_s": 1.1107993486803025e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 234800, "loss": 0.5876136515289545, "lr": 0.0003, "grad_norm": 0.1463273286950648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34368284786, "tokens_per_second": 149696.79550707154, "data_wait_s": 1.6377998690586537e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 234900, "loss": 0.5683729182928801, "lr": 0.0003, "grad_norm": 0.15020179548666457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34382846136, "tokens_per_second": 146679.03862138622, "data_wait_s": 2.205999771831557e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21403be270dc34614299aa762ec05883d846cdf2bc512722fe95ee0a6eef8331
-size 848610
+oid sha256:bd09fe952a0f13ac10820f24bfd3de15adb1455859b41f8d03d626ed4c4ce69a
+size 850728