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": 76165.90640865364,
-      "required_pages_per_second_upper_estimate": 0.006601867726777169
+      "required_token_positions_per_second": 76002.3518024707,
+      "required_pages_per_second_upper_estimate": 0.006587691228038591
     },
     {
       "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": 9520.738301081705,
-      "required_pages_per_second_upper_estimate": 0.0037979648560242957
+      "required_token_positions_per_second": 9500.293975308838,
+      "required_pages_per_second_upper_estimate": 0.0037898093088035894
     },
     {
       "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": 19041.47660216341,
-      "required_pages_per_second_upper_estimate": 0.004404910493368026
+      "required_token_positions_per_second": 19000.587950617675,
+      "required_pages_per_second_upper_estimate": 0.0043954516234486095
     },
     {
       "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": 28562.214903245116,
-      "required_pages_per_second_upper_estimate": 0.0017392890332473855
+      "required_token_positions_per_second": 28500.88192592651,
+      "required_pages_per_second_upper_estimate": 0.0017355541767179452
     },
     {
       "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": 28562.214903245116,
-      "required_pages_per_second_upper_estimate": 0.0021987983685230637
+      "required_token_positions_per_second": 28500.88192592651,
+      "required_pages_per_second_upper_estimate": 0.00219407678614853
     },
     {
       "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": 28562.214903245116,
-      "required_pages_per_second_upper_estimate": 0.35464711751424954
+      "required_token_positions_per_second": 28500.88192592651,
+      "required_pages_per_second_upper_estimate": 0.35388556720422304
     }
   ],
   "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": 0.08015877382023007,
-    "producer_p95_seconds": 0.21611261699945317,
-    "gpu_update_seconds": 0.967992156548712,
-    "producer_batches_per_second": 12.47524072963832,
-    "gpu_batches_per_second": 1.033066221905567,
-    "producer_headroom": 12.075935177346924,
+    "producer_batch_seconds": 0.2581904432109212,
+    "producer_p95_seconds": 0.2337775319992943,
+    "gpu_update_seconds": 0.9700752444032033,
+    "producer_batches_per_second": 3.8731100483958616,
+    "gpu_batches_per_second": 1.0308478705847264,
+    "producer_headroom": 3.757208176798118,
     "sustainable": true,
-    "buffer_seconds": 60.015513706020144,
+    "buffer_seconds": 60.1446651529986,
     "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": 502623.33866337617,
-    "upload_bytes_per_second": 5205137.642691322,
+    "generation_bytes_per_second": 448433.2522006597,
+    "upload_bytes_per_second": 5070975.459308068,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 107.85126800466503,
+    "estimated_full_shards_per_hour": 96.22333693041652,
     "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": 1.6520775649996722,
-      "bytes": 17392082,
-      "pages": 43,
-      "shards": 5
-    },
     {
       "seconds": 1.2575243370001772,
       "bytes": 1055853,
@@ -562,6 +556,12 @@
       "bytes": 7715646,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.6021141640012502,
+      "bytes": 6045584,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 169100, "loss": 0.6279681622982025, "lr": 0.0003, "grad_norm": 0.12974266857108036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24748531140, "tokens_per_second": 150551.60547993824, "data_wait_s": 1.5487999917240813e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 169200, "loss": 0.5250497795641422, "lr": 0.0003, "grad_norm": 0.1483682590171452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24763144161, "tokens_per_second": 150077.83161036504, "data_wait_s": 8.96399978955742e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 169300, "loss": 0.6024130657315254, "lr": 0.0003, "grad_norm": 0.15711487218539658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24777769244, "tokens_per_second": 149740.9886463385, "data_wait_s": 1.4848999853711575e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 169400, "loss": 0.6082226745784283, "lr": 0.0003, "grad_norm": 0.17083282993373824, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24792338896, "tokens_per_second": 148425.14187539616, "data_wait_s": 1.3553000826505013e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d17a75ea74ec522fdd1ae838a3b61e9e45421b23980a064c8b2029fa619b70a1
-size 304605
+oid sha256:4ba67380fb601df2cf2480a38bee568f6c071634fffbb9278a12727e235df265
+size 306723