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": 75538.99012791507,
-      "required_pages_per_second_upper_estimate": 0.006547528212467536
+      "required_token_positions_per_second": 73833.13204758412,
+      "required_pages_per_second_upper_estimate": 0.006399668757522228
     },
     {
       "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": 9442.373765989383,
-      "required_pages_per_second_upper_estimate": 0.003766704071321758
+      "required_token_positions_per_second": 9229.141505948015,
+      "required_pages_per_second_upper_estimate": 0.0036816425346848634
     },
     {
       "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": 18884.747531978766,
-      "required_pages_per_second_upper_estimate": 0.004368653981318235
+      "required_token_positions_per_second": 18458.28301189603,
+      "required_pages_per_second_upper_estimate": 0.004269998920116302
     },
     {
       "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": 28327.12129796815,
-      "required_pages_per_second_upper_estimate": 0.0017249730661268398
+      "required_token_positions_per_second": 27687.424517844043,
+      "required_pages_per_second_upper_estimate": 0.0016860188884469038
     },
     {
       "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": 28327.12129796815,
-      "required_pages_per_second_upper_estimate": 0.002180700212008092
+      "required_token_positions_per_second": 27687.424517844043,
+      "required_pages_per_second_upper_estimate": 0.0021314545830801146
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28327.12129796815,
-      "required_pages_per_second_upper_estimate": 0.35172804174439265
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27687.424517844043,
+      "required_pages_per_second_upper_estimate": 0.3437851486626525
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.06506710257053783,
-    "producer_p95_seconds": 0.2081564639993303,
-    "gpu_update_seconds": 0.9760257567006336,
-    "producer_batches_per_second": 15.368749498503053,
-    "gpu_batches_per_second": 1.0245631256498897,
-    "producer_headroom": 15.000295358818926,
+    "producer_batch_seconds": 0.37525666853136386,
+    "producer_p95_seconds": 0.3046484290025546,
+    "gpu_update_seconds": 0.9985760857670732,
+    "producer_batches_per_second": 2.6648427166229567,
+    "gpu_batches_per_second": 1.0014259446558176,
+    "producer_headroom": 2.661048209150246,
     "sustainable": true,
-    "buffer_seconds": 60.51359691543929,
+    "buffer_seconds": 60.91314123179147,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 278024.65559413296,
+    "generation_bytes_per_second": 269763.1950518941,
     "upload_bytes_per_second": 5136881.452576272,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.65761900775902,
+    "estimated_full_shards_per_hour": 57.88490189235323,
     "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-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -186,3 +186,4 @@
 {"step": 255700, "loss": 0.6096172984689474, "lr": 0.0003, "grad_norm": 0.17134667809357576, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37427631977, "tokens_per_second": 150449.1907204958, "data_wait_s": 1.467999754822813e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255800, "loss": 0.6042794194072485, "lr": 0.0003, "grad_norm": 0.14838269205064727, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37442394541, "tokens_per_second": 140723.64876658912, "data_wait_s": 1.5637997421436012e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 255900, "loss": 0.6159732881933451, "lr": 0.0003, "grad_norm": 0.1648850653482918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37457210876, "tokens_per_second": 151658.94846547183, "data_wait_s": 1.2847998732468113e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 256000, "loss": 0.5724415425211191, "lr": 0.0003, "grad_norm": 0.15820589177725958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 37471774801, "tokens_per_second": 147479.50363534992, "data_wait_s": 1.5329998859670013e-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 sha256:163b5b6ce5d211801dff7501cfd989bc2aa650102f8fb4d30fb541a28304e680
-size 401481
+oid sha256:e7409efbfc21d29a97cf71171ef1c97f415a933473f41321f57d99a8aa3a1e2c
+size 403599