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": 74343.7919209492,
-      "required_pages_per_second_upper_estimate": 0.006443931461089896
+      "required_token_positions_per_second": 75402.45433018998,
+      "required_pages_per_second_upper_estimate": 0.006535693635567542
     },
     {
       "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": 9292.97399011865,
-      "required_pages_per_second_upper_estimate": 0.0037071062670012167
+      "required_token_positions_per_second": 9425.306791273748,
+      "required_pages_per_second_upper_estimate": 0.0037598957999336797
     },
     {
       "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": 18585.9479802373,
-      "required_pages_per_second_upper_estimate": 0.004299531963715351
+      "required_token_positions_per_second": 18850.613582547496,
+      "required_pages_per_second_upper_estimate": 0.004360757692854303
     },
     {
       "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": 27878.92197035595,
-      "required_pages_per_second_upper_estimate": 0.0016976800785953937
+      "required_token_positions_per_second": 28275.92037382124,
+      "required_pages_per_second_upper_estimate": 0.0017218551984767812
     },
     {
       "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": 27878.92197035595,
-      "required_pages_per_second_upper_estimate": 0.0021461965870733578
+      "required_token_positions_per_second": 28275.92037382124,
+      "required_pages_per_second_upper_estimate": 0.0021767586231340335
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27878.92197035595,
-      "required_pages_per_second_upper_estimate": 0.34616290612210476
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28275.92037382124,
+      "required_pages_per_second_upper_estimate": 0.35109229762495797
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.0678645764843182,
-    "producer_p95_seconds": 0.16497391799930483,
-    "gpu_update_seconds": 0.9917169691639622,
-    "producer_batches_per_second": 14.735227887719521,
-    "gpu_batches_per_second": 1.0083522124694715,
-    "producer_headroom": 14.613175540749497,
+    "producer_batch_seconds": 0.1862787745702903,
+    "producer_p95_seconds": 0.1637777689993527,
+    "gpu_update_seconds": 0.9777931057408624,
+    "producer_batches_per_second": 5.368298145115083,
+    "gpu_batches_per_second": 1.0227112403725855,
+    "producer_headroom": 5.249084915854987,
     "sustainable": true,
-    "buffer_seconds": 60.494735119001696,
+    "buffer_seconds": 60.62317255593347,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369332.29578550963,
-    "upload_bytes_per_second": 4987128.874655885,
+    "generation_bytes_per_second": 369141.59805309924,
+    "upload_bytes_per_second": 5031288.450927107,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.25011305974928,
+    "estimated_full_shards_per_hour": 79.20919376559003,
     "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": [
@@ -238,6 +238,12 @@
       "bytes": 2621863,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0488426049996633,
+      "bytes": 5758950,
+      "pages": 26,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -36,3 +36,4 @@
 {"step": 358600, "loss": 0.5974908228963614, "lr": 0.0003, "grad_norm": 0.16560434333579174, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52489362573, "tokens_per_second": 149132.2637041259, "data_wait_s": 6.9150000854278915e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 358700, "loss": 0.5217609908431768, "lr": 0.0003, "grad_norm": 0.17284499424465385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52503965450, "tokens_per_second": 148584.709432047, "data_wait_s": 1.101300040318165e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 358800, "loss": 0.5568342115730047, "lr": 0.0003, "grad_norm": 0.15941198218727748, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52518633527, "tokens_per_second": 149875.30207616428, "data_wait_s": 1.5557999176962767e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 358900, "loss": 0.5688133426010609, "lr": 0.0003, "grad_norm": 0.15956965421208585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52533196374, "tokens_per_second": 147767.6676919024, "data_wait_s": 8.003999937500339e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47fdf81059da928fd2e2173aabcc24faa9a778f20a6ea4703e5d7cbb63fef756
-size 80649
+oid sha256:4c1e28ff20c951108abd3aa25b7466dfa2ccc4a569e12ef3666f197f79bdf6ab
+size 82767