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": 75298.41122862679,
-      "required_pages_per_second_upper_estimate": 0.006526675443218875
+      "required_token_positions_per_second": 75757.22172089446,
+      "required_pages_per_second_upper_estimate": 0.006566443973844075
     },
     {
       "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": 9412.301403578349,
-      "required_pages_per_second_upper_estimate": 0.0037547077563341107
+      "required_token_positions_per_second": 9469.652715111808,
+      "required_pages_per_second_upper_estimate": 0.0037775860519833285
     },
     {
       "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": 18824.602807156698,
-      "required_pages_per_second_upper_estimate": 0.004354740557741654
+      "required_token_positions_per_second": 18939.305430223616,
+      "required_pages_per_second_upper_estimate": 0.004381274990890967
     },
     {
       "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": 28236.904210735047,
-      "required_pages_per_second_upper_estimate": 0.0017194793188468195
+      "required_token_positions_per_second": 28408.958145335426,
+      "required_pages_per_second_upper_estimate": 0.0017299565007667518
     },
     {
       "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": 28236.904210735047,
-      "required_pages_per_second_upper_estimate": 0.002173755050896006
+      "required_token_positions_per_second": 28408.958145335426,
+      "required_pages_per_second_upper_estimate": 0.002187000239057317
     },
     {
       "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": 28236.904210735047,
-      "required_pages_per_second_upper_estimate": 0.3506078474581254
+      "required_token_positions_per_second": 28408.958145335426,
+      "required_pages_per_second_upper_estimate": 0.35274418149838493
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.1210159916873863,
-    "producer_p95_seconds": 0.3887702639985946,
-    "gpu_update_seconds": 0.979144165155642,
-    "producer_batches_per_second": 8.26337070048761,
-    "gpu_batches_per_second": 1.0213000654924425,
-    "producer_headroom": 8.091031205900531,
+    "producer_batch_seconds": 0.14332634784364018,
+    "producer_p95_seconds": 0.38965308600018034,
+    "gpu_update_seconds": 0.9732141481062948,
+    "producer_batches_per_second": 6.9770842210459145,
+    "gpu_batches_per_second": 1.0275230810668194,
+    "producer_headroom": 6.790197076451071,
     "sustainable": true,
-    "buffer_seconds": 60.7069382396498,
+    "buffer_seconds": 60.33927718259028,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 391826.58073969005,
+    "generation_bytes_per_second": 430048.5245690244,
     "upload_bytes_per_second": 4244394.412710249,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 84.0768629707625,
+    "estimated_full_shards_per_hour": 92.27840235522316,
     "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": [

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -338,3 +338,4 @@
 {"step": 33800, "loss": 0.6564355380833149, "lr": 0.0003, "grad_norm": 0.13291179917537557, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4946206923, "tokens_per_second": 148956.07547159746, "data_wait_s": 1.4435994671657681e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 33900, "loss": 0.6282978411763906, "lr": 0.0003, "grad_norm": 0.13631934542786595, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4960938164, "tokens_per_second": 149625.6474962816, "data_wait_s": 1.1384996469132602e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 34000, "loss": 0.6324362829327583, "lr": 0.0003, "grad_norm": 0.1366603905928768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4975444748, "tokens_per_second": 148308.1229782295, "data_wait_s": 1.0525000107008964e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 34100, "loss": 0.6961428821086884, "lr": 0.0003, "grad_norm": 0.13530116001377898, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 4989998423, "tokens_per_second": 148800.8387634716, "data_wait_s": 9.483999747317284e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:387e877cf1bd9920de35a0810e80c841c4a03fc748ba88967e706469f9733db9
-size 719209
+oid sha256:e7bf3f0da5bedb6dbbdd567f39454525a74dc56e10fde3050bb92859e29d0bc3
+size 721327