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": 75054.36740070669,
-      "required_pages_per_second_upper_estimate": 0.0065055223427381865
+      "required_token_positions_per_second": 75121.23398537125,
+      "required_pages_per_second_upper_estimate": 0.006511318168825102
     },
     {
       "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": 9381.795925088336,
-      "required_pages_per_second_upper_estimate": 0.0037425386648668965
+      "required_token_positions_per_second": 9390.154248171406,
+      "required_pages_per_second_upper_estimate": 0.0037458729249127995
     },
     {
       "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": 18763.591850176672,
-      "required_pages_per_second_upper_estimate": 0.004340626746600494
+      "required_token_positions_per_second": 18780.308496342812,
+      "required_pages_per_second_upper_estimate": 0.004344493848488107
     },
     {
       "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": 28145.38777526501,
-      "required_pages_per_second_upper_estimate": 0.0017139064480763233
+      "required_token_positions_per_second": 28170.462744514218,
+      "required_pages_per_second_upper_estimate": 0.0017154333821454036
     },
     {
       "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": 28145.38777526501,
-      "required_pages_per_second_upper_estimate": 0.0021667098623598152
+      "required_token_positions_per_second": 28170.462744514218,
+      "required_pages_per_second_upper_estimate": 0.0021686402029045767
     },
     {
       "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": 28145.38777526501,
-      "required_pages_per_second_upper_estimate": 0.3494715196153943
+      "required_token_positions_per_second": 28170.462744514218,
+      "required_pages_per_second_upper_estimate": 0.34978286681294585
     }
   ],
   "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": 0.44149891271115393,
-    "producer_p95_seconds": 0.41396072399948025,
-    "gpu_update_seconds": 0.9823279117972527,
-    "producer_batches_per_second": 2.2650112405922043,
-    "gpu_batches_per_second": 1.0179900092326755,
-    "producer_headroom": 2.2249837621682444,
+    "producer_batch_seconds": 0.10366545646087388,
+    "producer_p95_seconds": 0.4013702709999052,
+    "gpu_update_seconds": 0.9814535263672245,
+    "producer_batches_per_second": 9.646414863155758,
+    "gpu_batches_per_second": 1.0188969453310988,
+    "producer_headroom": 9.467507884245427,
     "sustainable": true,
-    "buffer_seconds": 60.904330531429665,
+    "buffer_seconds": 60.85011863476792,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319312.0879352497,
+    "generation_bytes_per_second": 319177.3490550403,
     "upload_bytes_per_second": 5822128.923463643,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.51694086592788,
+    "estimated_full_shards_per_hour": 68.48802903879553,
     "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-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -376,3 +376,4 @@
 {"step": 232600, "loss": 0.5765535496175289, "lr": 0.0003, "grad_norm": 0.16614487692741248, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34046224255, "tokens_per_second": 137174.81873389723, "data_wait_s": 8.55699909152463e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 232700, "loss": 0.5782700162380934, "lr": 0.0003, "grad_norm": 0.15028562898498768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34060838035, "tokens_per_second": 147250.36277060804, "data_wait_s": 2.0628998754546046e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 232800, "loss": 0.6391399744898081, "lr": 0.0003, "grad_norm": 0.2039060200830334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34075313299, "tokens_per_second": 146633.77329801678, "data_wait_s": 1.0544004908297211e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 232900, "loss": 0.5621838551014662, "lr": 0.0003, "grad_norm": 0.1732460337275868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34089887453, "tokens_per_second": 147968.77266469318, "data_wait_s": 1.0024996299762279e-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:df693841fa44d4c24d4c74c8e7af3eed158fbbdba9a84635ded49bdbc9cccf8c
-size 806250
+oid sha256:f20d5571052f214b9142f15afc54eb2e90782d9eef33f54ef1ad1e9cb0791bdc
+size 808368