Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77513.03011955845,
-      "required_pages_per_second_upper_estimate": 0.005362531278680116
+      "required_token_positions_per_second": 77569.0575284993,
+      "required_pages_per_second_upper_estimate": 0.005366407384832143
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9689.128764944806,
-      "required_pages_per_second_upper_estimate": 0.004056635881865838
+      "required_token_positions_per_second": 9696.132191062412,
+      "required_pages_per_second_upper_estimate": 0.0040595680701331115
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19378.25752988961,
-      "required_pages_per_second_upper_estimate": 0.004035980004459058
+      "required_token_positions_per_second": 19392.264382124824,
+      "required_pages_per_second_upper_estimate": 0.004038897262394119
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29067.386294834414,
-      "required_pages_per_second_upper_estimate": 0.001742545425046329
+      "required_token_positions_per_second": 29088.396573187234,
+      "required_pages_per_second_upper_estimate": 0.0017438049591527432
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29067.386294834414,
-      "required_pages_per_second_upper_estimate": 0.0022598614305030834
+      "required_token_positions_per_second": 29088.396573187234,
+      "required_pages_per_second_upper_estimate": 0.0022614948872305674
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29067.386294834414,
-      "required_pages_per_second_upper_estimate": 0.3728452212623544
+      "required_token_positions_per_second": 29088.396573187234,
+      "required_pages_per_second_upper_estimate": 0.37311471855398515
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.2895324169296032,
-    "producer_p95_seconds": 0.3094242799998028,
-    "gpu_update_seconds": 0.9511691116484506,
-    "producer_batches_per_second": 3.4538446872535853,
-    "gpu_batches_per_second": 1.0513377566129345,
-    "producer_headroom": 3.2851903829467135,
+    "producer_batch_seconds": 0.10406298535161795,
+    "producer_p95_seconds": 0.3004552580023301,
+    "gpu_update_seconds": 0.9504820910439956,
+    "producer_batches_per_second": 9.609564790219158,
+    "gpu_batches_per_second": 1.0520976769815986,
+    "producer_headroom": 9.133719235830261,
     "sustainable": true,
-    "buffer_seconds": 60.87482314550084,
+    "buffer_seconds": 60.83085382681572,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 40938.33103456794,
+    "generation_bytes_per_second": 40495.98194763162,
     "upload_bytes_per_second": 5319616.870726861,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 8.78441284444598,
+    "estimated_full_shards_per_hour": 8.689495027749171,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -333,3 +333,4 @@
 {"step": 718300, "loss": 0.5473929643630981, "lr": 8e-05, "grad_norm": 0.21766812750350945, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105139523634, "tokens_per_second": 151927.6906970336, "data_wait_s": 1.7646001651883125e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 718400, "loss": 0.5698359608650208, "lr": 8e-05, "grad_norm": 0.2282269000352339, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105153987586, "tokens_per_second": 151202.6121852039, "data_wait_s": 1.5789002645760775e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 718500, "loss": 0.5605199337005615, "lr": 8e-05, "grad_norm": 0.23433685480020014, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105168631666, "tokens_per_second": 153220.88380636225, "data_wait_s": 1.1659998563118279e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 718600, "loss": 0.5746077299118042, "lr": 8e-05, "grad_norm": 0.22633490972812828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 105183270447, "tokens_per_second": 136851.05222668315, "data_wait_s": 1.6204998246394098e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:39fa6047eed6dd7a7f23b9daf320fa6b5113e9d11782081d6c5b1d1451edbf98
-size 714393
+oid sha256:b3b81fbf36a2bf185bd4bddcd01b7f87ab78912be99d016c87aa5fe58a70e1b8
+size 716511