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": 74960.60535150075,
-      "required_pages_per_second_upper_estimate": 0.006497395285950757
+      "required_token_positions_per_second": 74564.04182435619,
+      "required_pages_per_second_upper_estimate": 0.006463022164498938
     },
     {
       "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": 9370.075668937594,
-      "required_pages_per_second_upper_estimate": 0.0037378632794549203
+      "required_token_positions_per_second": 9320.505228044523,
+      "required_pages_per_second_upper_estimate": 0.0037180888894385367
     },
     {
       "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": 18740.151337875188,
-      "required_pages_per_second_upper_estimate": 0.004335204196618473
+      "required_token_positions_per_second": 18641.010456089047,
+      "required_pages_per_second_upper_estimate": 0.004312269698437176
     },
     {
       "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": 28110.22700681278,
-      "required_pages_per_second_upper_estimate": 0.0017117653417518775
+      "required_token_positions_per_second": 27961.51568413357,
+      "required_pages_per_second_upper_estimate": 0.0017027096024287245
     },
     {
       "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": 28110.22700681278,
-      "required_pages_per_second_upper_estimate": 0.002164003089073126
+      "required_token_positions_per_second": 27961.51568413357,
+      "required_pages_per_second_upper_estimate": 0.0021525548797939917
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28110.22700681278,
-      "required_pages_per_second_upper_estimate": 0.3490349405467398
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27961.51568413357,
+      "required_pages_per_second_upper_estimate": 0.34718844362384454
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1132801141251889,
-    "producer_p95_seconds": 0.3917120930054807,
-    "gpu_update_seconds": 0.9835566249002274,
-    "producer_batches_per_second": 8.827674722280674,
-    "gpu_batches_per_second": 1.016718280049652,
-    "producer_headroom": 8.682517955563432,
+    "producer_batch_seconds": 0.2582017903126257,
+    "producer_p95_seconds": 0.3919388270005584,
+    "gpu_update_seconds": 0.9887876005122472,
+    "producer_batches_per_second": 3.87293983821421,
+    "gpu_batches_per_second": 1.0113395429735812,
+    "producer_headroom": 3.82951488955612,
     "sustainable": true,
-    "buffer_seconds": 60.9805107438141,
+    "buffer_seconds": 60.316043631247084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 162,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 687288.858706644,
-    "upload_bytes_per_second": 5332277.649329477,
+    "generation_bytes_per_second": 623451.2923197498,
+    "upload_bytes_per_second": 5317014.665125242,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.73809490632767,
+    "estimated_full_shards_per_hour": 66.88906706425844,
     "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": 1.168987963003019,
-      "bytes": 6662663,
-      "pages": 33,
-      "shards": 3
-    },
     {
       "seconds": 3.650391375998879,
       "bytes": 6163332,
@@ -562,6 +556,12 @@
       "bytes": 1028230,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2517391190049239,
+      "bytes": 5921573,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -329,3 +329,4 @@
 {"step": 227900, "loss": 0.5573900323361158, "lr": 0.0003, "grad_norm": 0.15145536798889647, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33357920288, "tokens_per_second": 149036.76707716731, "data_wait_s": 1.0244999430142343e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 228000, "loss": 0.611876580864191, "lr": 0.0003, "grad_norm": 0.15276256277388992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33372666072, "tokens_per_second": 149702.1373013235, "data_wait_s": 8.10199708212167e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 228100, "loss": 0.5718318037688732, "lr": 0.0003, "grad_norm": 0.15779332785902814, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33387338896, "tokens_per_second": 148749.5906140079, "data_wait_s": 2.1855004888493568e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 228200, "loss": 0.5915424004197121, "lr": 0.0003, "grad_norm": 0.15198145500474589, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33401767184, "tokens_per_second": 147298.9726504159, "data_wait_s": 1.1588002962525934e-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:a349a9d0f3278a690009188e096778e34e8bc32e726090c6214972398338fa69
-size 705921
+oid sha256:c4e862a11884833a01768514f10d28224171eaae0203866902eed1bc7a2e73ff
+size 708039