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": 75960.48551618592,
-      "required_pages_per_second_upper_estimate": 0.006584062364452558
+      "required_token_positions_per_second": 75472.91571123064,
+      "required_pages_per_second_upper_estimate": 0.006541801049493408
     },
     {
       "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": 9495.06068952324,
-      "required_pages_per_second_upper_estimate": 0.003787721672859119
+      "required_token_positions_per_second": 9434.11446390383,
+      "required_pages_per_second_upper_estimate": 0.0037634093122322606
     },
     {
       "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": 18990.12137904648,
-      "required_pages_per_second_upper_estimate": 0.004393030366320981
+      "required_token_positions_per_second": 18868.22892780766,
+      "required_pages_per_second_upper_estimate": 0.004364832692960754
     },
     {
       "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": 28485.182068569717,
-      "required_pages_per_second_upper_estimate": 0.001734598137775702
+      "required_token_positions_per_second": 28302.34339171149,
+      "required_pages_per_second_upper_estimate": 0.0017234642216354364
     },
     {
       "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": 28485.182068569717,
-      "required_pages_per_second_upper_estimate": 0.002192868167669217
+      "required_token_positions_per_second": 28302.34339171149,
+      "required_pages_per_second_upper_estimate": 0.0021787927401948196
     },
     {
       "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": 28485.182068569717,
-      "required_pages_per_second_upper_estimate": 0.35369062751989416
+      "required_token_positions_per_second": 28302.34339171149,
+      "required_pages_per_second_upper_estimate": 0.3514203830750027
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.050354257578121064,
-    "producer_p95_seconds": 0.17949672999930044,
-    "gpu_update_seconds": 0.9706099098628035,
-    "producer_batches_per_second": 19.859293892846516,
-    "gpu_batches_per_second": 1.0302800227347264,
-    "producer_headroom": 19.27562745527468,
+    "producer_batch_seconds": 0.37397229494531103,
+    "producer_p95_seconds": 0.20154814399938914,
+    "gpu_update_seconds": 0.9768802398213033,
+    "producer_batches_per_second": 2.6739948747974447,
+    "gpu_batches_per_second": 1.0236669340173428,
+    "producer_headroom": 2.6121727545730637,
     "sustainable": true,
-    "buffer_seconds": 60.17781441149381,
+    "buffer_seconds": 60.56657486892081,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401738.40105100675,
-    "upload_bytes_per_second": 3271094.7586607332,
+    "generation_bytes_per_second": 372893.2640622187,
+    "upload_bytes_per_second": 3237772.7279057847,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.20370887420322,
+    "estimated_full_shards_per_hour": 80.01421395683214,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 7809668,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.2807725910006411,
+      "bytes": 1082933,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -56,3 +56,4 @@
 {"step": 242700, "loss": 0.6064958516508341, "lr": 0.0003, "grad_norm": 0.13906499441895614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35523573545, "tokens_per_second": 148033.97377287998, "data_wait_s": 1.595899993844796e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 242800, "loss": 0.5889896117150784, "lr": 0.0003, "grad_norm": 0.15594127761860646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35538245263, "tokens_per_second": 149761.95275660782, "data_wait_s": 9.946000318450388e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 242900, "loss": 0.5998102594166994, "lr": 0.0003, "grad_norm": 0.13357588322711464, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35552906976, "tokens_per_second": 149279.46177723623, "data_wait_s": 1.0071999895444605e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 243000, "loss": 0.5930889509618282, "lr": 0.0003, "grad_norm": 0.1354303855648867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 35567657221, "tokens_per_second": 150613.2766396233, "data_wait_s": 1.6164000044227578e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:21f7dac9f838ecd91614127b7d7fe9fd103183aaf2ae397d5a832bb64cab9277
-size 123792
+oid sha256:809fe53e91ee501e8008793b1c113837bf6d9fdd70270c664d06900a98afc008
+size 125910