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": 76025.95885552705,
-      "required_pages_per_second_upper_estimate": 0.006589737427565986
+      "required_token_positions_per_second": 75793.62418405307,
+      "required_pages_per_second_upper_estimate": 0.006569599247089463
     },
     {
       "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": 9503.24485694088,
-      "required_pages_per_second_upper_estimate": 0.0037909864596062234
+      "required_token_positions_per_second": 9474.203023006634,
+      "required_pages_per_second_upper_estimate": 0.00377940123783574
     },
     {
       "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": 19006.48971388176,
-      "required_pages_per_second_upper_estimate": 0.00439681689251227
+      "required_token_positions_per_second": 18948.406046013268,
+      "required_pages_per_second_upper_estimate": 0.004383380258188535
     },
     {
       "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": 28509.734570822642,
-      "required_pages_per_second_upper_estimate": 0.0017360932563458813
+      "required_token_positions_per_second": 28422.6090690199,
+      "required_pages_per_second_upper_estimate": 0.001730787770398275
     },
     {
       "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": 28509.734570822642,
-      "required_pages_per_second_upper_estimate": 0.0021947582872583237
+      "required_token_positions_per_second": 28422.6090690199,
+      "required_pages_per_second_upper_estimate": 0.0021880511249507233
     },
     {
       "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": 28509.734570822642,
-      "required_pages_per_second_upper_estimate": 0.35399548742593645
+      "required_token_positions_per_second": 28422.6090690199,
+      "required_pages_per_second_upper_estimate": 0.35291368028384346
     }
   ],
   "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": 0.09140126632829038,
-    "producer_p95_seconds": 0.2975480029999744,
-    "gpu_update_seconds": 0.9697740233714923,
-    "producer_batches_per_second": 10.940767455105615,
-    "gpu_batches_per_second": 1.0311680617340366,
-    "producer_headroom": 10.610072073709654,
+    "producer_batch_seconds": 0.1301028783750553,
+    "producer_p95_seconds": 0.297439073001442,
+    "gpu_update_seconds": 0.9727467289459992,
+    "producer_batches_per_second": 7.686225028144578,
+    "gpu_batches_per_second": 1.0280168210727683,
+    "producer_headroom": 7.476750254070509,
     "sustainable": true,
-    "buffer_seconds": 60.125989449032524,
+    "buffer_seconds": 60.31029719465195,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357338.52778235904,
-    "upload_bytes_per_second": 2231184.142501255,
+    "generation_bytes_per_second": 365004.6061232208,
+    "upload_bytes_per_second": 2254259.001417754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.67652964690284,
+    "estimated_full_shards_per_hour": 78.3214916016814,
     "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": 1.1701329820025421,
-      "bytes": 6128160,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9900568350021786,
       "bytes": 2575859,
@@ -562,6 +556,12 @@
       "bytes": 1197583,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.2053027080000902,
+      "bytes": 10494285,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -180,3 +180,4 @@
 {"step": 298200, "loss": 0.5542546007782221, "lr": 0.0003, "grad_norm": 0.14852502627466904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43648809275, "tokens_per_second": 148614.5832521336, "data_wait_s": 1.6843998309923336e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 298300, "loss": 0.6011155806481838, "lr": 0.0003, "grad_norm": 0.15396914818704283, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43663407844, "tokens_per_second": 149417.20726141307, "data_wait_s": 9.9300013971515e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 298400, "loss": 0.6268672812730074, "lr": 0.0003, "grad_norm": 0.14880373418193077, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43677955154, "tokens_per_second": 149101.95634820187, "data_wait_s": 1.704800160950981e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 298500, "loss": 0.5878512840718031, "lr": 0.0003, "grad_norm": 0.16763574281207322, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 43692508028, "tokens_per_second": 149412.37226701714, "data_wait_s": 1.4777997421333566e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d290d92298661351e8a5e0ef73dc44ec73de1a72e2739867a82545e78303e67d
-size 387990
+oid sha256:51c845cff97b55bb92843b77a1a764b64ad5f02f4d1f1d185d9793f74192f525
+size 390108