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": 75176.72305627036,
-      "required_pages_per_second_upper_estimate": 0.006516127820854872
+      "required_token_positions_per_second": 75400.62889578588,
+      "required_pages_per_second_upper_estimate": 0.006535535411539915
     },
     {
       "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": 9397.090382033795,
-      "required_pages_per_second_upper_estimate": 0.0037486398524149496
+      "required_token_positions_per_second": 9425.078611973235,
+      "required_pages_per_second_upper_estimate": 0.00375980477579912
     },
     {
       "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": 18794.18076406759,
-      "required_pages_per_second_upper_estimate": 0.004347702953482661
+      "required_token_positions_per_second": 18850.15722394647,
+      "required_pages_per_second_upper_estimate": 0.004360652122323593
     },
     {
       "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": 28191.27114610139,
-      "required_pages_per_second_upper_estimate": 0.0017167005046288166
+      "required_token_positions_per_second": 28275.235835919706,
+      "required_pages_per_second_upper_estimate": 0.0017218135137100706
     },
     {
       "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": 28191.27114610139,
-      "required_pages_per_second_upper_estimate": 0.0021702420912600995
+      "required_token_positions_per_second": 28275.235835919706,
+      "required_pages_per_second_upper_estimate": 0.002176705925511451
     },
     {
       "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": 28191.27114610139,
-      "required_pages_per_second_upper_estimate": 0.3500412375194183
+      "required_token_positions_per_second": 28275.235835919706,
+      "required_pages_per_second_upper_estimate": 0.3510837979552219
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.10076894139845649,
-    "producer_p95_seconds": 0.30415311099932296,
-    "gpu_update_seconds": 0.9807291007459052,
-    "producer_batches_per_second": 9.923692619195435,
-    "gpu_batches_per_second": 1.0196495640227643,
-    "producer_headroom": 9.732454138502316,
+    "producer_batch_seconds": 0.0945624471096096,
+    "producer_p95_seconds": 0.29891200799829676,
+    "gpu_update_seconds": 0.9778167779197481,
+    "producer_batches_per_second": 10.575022438250526,
+    "gpu_batches_per_second": 1.0226864813339014,
+    "producer_headroom": 10.340434366999167,
     "sustainable": true,
-    "buffer_seconds": 60.80520424624613,
+    "buffer_seconds": 60.62464023102438,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367744.86740680604,
-    "upload_bytes_per_second": 5499242.782645166,
+    "generation_bytes_per_second": 355958.19272034994,
+    "upload_bytes_per_second": 5580444.120847845,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.90948788312087,
+    "estimated_full_shards_per_hour": 76.38034187515139,
     "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": 1.2709880140009773,
-      "bytes": 1216081,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.5031630230005248,
       "bytes": 11828419,
@@ -562,6 +556,12 @@
       "bytes": 13355543,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.0210915280003974,
+      "bytes": 6074322,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -211,3 +211,4 @@
 {"step": 258200, "loss": 0.5067947395145893, "lr": 0.0003, "grad_norm": 0.15570042932146388, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37793466222, "tokens_per_second": 148696.57369766242, "data_wait_s": 1.4452001778408885e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 258300, "loss": 0.596388328820467, "lr": 0.0003, "grad_norm": 0.14077683984788145, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37807949727, "tokens_per_second": 146497.8689961445, "data_wait_s": 9.57899828790687e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 258400, "loss": 0.6657634973526001, "lr": 0.0003, "grad_norm": 0.18962240788548118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37822479157, "tokens_per_second": 147833.52644985594, "data_wait_s": 1.2987999070901424e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 258500, "loss": 0.6101230606436729, "lr": 0.0003, "grad_norm": 0.14810045976350694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37837180088, "tokens_per_second": 149825.3454071128, "data_wait_s": 1.1959997209487483e-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 sha256:26059da4a0cce111f3e3ec618460f3427d97b50be8a46272c16f758eeb9a6c00
-size 454431
+oid sha256:215b525f3b2012fa40d653eb6e2b6e9f054af7805b386a95fd3a22ea4990efb3
+size 456549