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": 74552.46993637337,
-      "required_pages_per_second_upper_estimate": 0.006462019142577268
+      "required_token_positions_per_second": 73963.59467393602,
+      "required_pages_per_second_upper_estimate": 0.006410976927319907
     },
     {
       "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": 9319.05874204667,
-      "required_pages_per_second_upper_estimate": 0.0037175118645471003
+      "required_token_positions_per_second": 9245.449334242003,
+      "required_pages_per_second_upper_estimate": 0.003688147971215096
     },
     {
       "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": 18638.11748409334,
-      "required_pages_per_second_upper_estimate": 0.004311600460280537
+      "required_token_positions_per_second": 18490.898668484006,
+      "required_pages_per_second_upper_estimate": 0.004277543978251979
     },
     {
       "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": 27957.176226140014,
-      "required_pages_per_second_upper_estimate": 0.0017024453522042934
+      "required_token_positions_per_second": 27736.34800272601,
+      "required_pages_per_second_upper_estimate": 0.0016889980719945277
     },
     {
       "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": 27957.176226140014,
-      "required_pages_per_second_upper_estimate": 0.0021522208163052593
+      "required_token_positions_per_second": 27736.34800272601,
+      "required_pages_per_second_upper_estimate": 0.0021352208483752023
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27957.176226140014,
-      "required_pages_per_second_upper_estimate": 0.3471345620787963
+      "required_token_positions_per_second": 27736.34800272601,
+      "required_pages_per_second_upper_estimate": 0.34439261460851545
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.44542471855470467,
-    "producer_p95_seconds": 0.333499052998377,
-    "gpu_update_seconds": 0.9889410781818897,
-    "producer_batches_per_second": 2.245048283904759,
-    "gpu_batches_per_second": 1.0111825891977726,
-    "producer_headroom": 2.220220470455174,
+    "producer_batch_seconds": 0.3750802613594999,
+    "producer_p95_seconds": 0.32386904600207345,
+    "gpu_update_seconds": 0.9968147211479562,
+    "producer_batches_per_second": 2.6660960413524366,
+    "gpu_batches_per_second": 1.003195457274523,
+    "producer_headroom": 2.657603782014399,
     "sustainable": true,
-    "buffer_seconds": 60.32540576909527,
+    "buffer_seconds": 60.80569799002533,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 364225.7773524917,
+    "generation_bytes_per_second": 343033.563595588,
     "upload_bytes_per_second": 4986447.894074069,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.15437307768882,
+    "estimated_full_shards_per_hour": 73.60701733494501,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -314,3 +314,4 @@
 {"step": 386400, "loss": 0.5703699178993702, "lr": 0.0003, "grad_norm": 0.18797698578785957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56558064033, "tokens_per_second": 149667.19114507549, "data_wait_s": 1.2093001714674756e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 386500, "loss": 0.5747383907437325, "lr": 0.0003, "grad_norm": 0.16812191695093157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56572755382, "tokens_per_second": 149876.05576409269, "data_wait_s": 9.35700154514052e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 386600, "loss": 0.6133938077837229, "lr": 0.0003, "grad_norm": 0.15190493584300963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56587505244, "tokens_per_second": 144101.1346329765, "data_wait_s": 8.521001291228458e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 386700, "loss": 0.6168648153543472, "lr": 0.0003, "grad_norm": 0.17218091571528626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56602285564, "tokens_per_second": 151670.51335805992, "data_wait_s": 9.251998562831432e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e7a9758d078e89f91f9f9dc9e0b9721dc5221d1a75dd17ca50fb12c492656ec2
-size 674151
+oid sha256:222af66e0bff9266c6052ae5d3f987362dbc5b92fcb5bc6761bbe63c8b1887f5
+size 676269