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": 75121.23398537125,
-      "required_pages_per_second_upper_estimate": 0.006511318168825102
+      "required_token_positions_per_second": 75239.88530379745,
+      "required_pages_per_second_upper_estimate": 0.006521602564387267
     },
     {
       "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": 9390.154248171406,
-      "required_pages_per_second_upper_estimate": 0.0037458729249127995
+      "required_token_positions_per_second": 9404.985662974681,
+      "required_pages_per_second_upper_estimate": 0.0037517893980272385
     },
     {
       "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": 18780.308496342812,
-      "required_pages_per_second_upper_estimate": 0.004344493848488107
+      "required_token_positions_per_second": 18809.971325949362,
+      "required_pages_per_second_upper_estimate": 0.0043513558220696115
     },
     {
       "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": 28170.462744514218,
-      "required_pages_per_second_upper_estimate": 0.0017154333821454036
+      "required_token_positions_per_second": 28214.956988924045,
+      "required_pages_per_second_upper_estimate": 0.0017181428481867031
     },
     {
       "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": 28170.462744514218,
-      "required_pages_per_second_upper_estimate": 0.0021686402029045767
+      "required_token_positions_per_second": 28214.956988924045,
+      "required_pages_per_second_upper_estimate": 0.0021720654930071966
     },
     {
       "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": 28170.462744514218,
-      "required_pages_per_second_upper_estimate": 0.34978286681294585
+      "required_token_positions_per_second": 28214.956988924045,
+      "required_pages_per_second_upper_estimate": 0.3503353364158591
     }
   ],
   "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": 0.10366545646087388,
-    "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,
+    "producer_batch_seconds": 0.11687535580415442,
+    "producer_p95_seconds": 0.39247806200000923,
+    "gpu_update_seconds": 0.9799057999930105,
+    "producer_batches_per_second": 8.556123685096445,
+    "gpu_batches_per_second": 1.0205062568331902,
+    "producer_headroom": 8.384195224483578,
     "sustainable": true,
-    "buffer_seconds": 60.85011863476792,
+    "buffer_seconds": 60.754159599566655,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 319177.3490550403,
-    "upload_bytes_per_second": 5822128.923463643,
+    "generation_bytes_per_second": 319110.0538364989,
+    "upload_bytes_per_second": 5810487.061690187,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.48802903879553,
+    "estimated_full_shards_per_hour": 68.47358905144905,
     "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": 0.7174145440003485,
-      "bytes": 276652,
-      "pages": 9,
-      "shards": 1
-    },
     {
       "seconds": 0.9002093449998938,
       "bytes": 1095526,
@@ -562,6 +556,12 @@
       "bytes": 17372933,
       "pages": 43,
       "shards": 5
+    },
+    {
+      "seconds": 0.9750086150015704,
+      "bytes": 928101,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -377,3 +377,4 @@
 {"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}
+{"step": 233000, "loss": 0.5900715999305248, "lr": 0.0003, "grad_norm": 0.16285978683707156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 34104661909, "tokens_per_second": 149594.95634956044, "data_wait_s": 1.3912002032157034e-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:f20d5571052f214b9142f15afc54eb2e90782d9eef33f54ef1ad1e9cb0791bdc
-size 808368
+oid sha256:4e773c0d09ea58e5f69181949b02f7d7027cb0f42b4cb99856510a4fc9fa0d42
+size 810486