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": 76111.26401445507,
-      "required_pages_per_second_upper_estimate": 0.0065971314625379186
+      "required_token_positions_per_second": 74231.25154427071,
+      "required_pages_per_second_upper_estimate": 0.006434176746470382
     },
     {
       "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": 9513.908001806883,
-      "required_pages_per_second_upper_estimate": 0.003795240147521495
+      "required_token_positions_per_second": 9278.906443033839,
+      "required_pages_per_second_upper_estimate": 0.003701494512140513
     },
     {
       "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": 19027.816003613767,
-      "required_pages_per_second_upper_estimate": 0.004401750354311889
+      "required_token_positions_per_second": 18557.812886067677,
+      "required_pages_per_second_upper_estimate": 0.004293023404839954
     },
     {
       "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": 28541.724005420652,
-      "required_pages_per_second_upper_estimate": 0.001738041245077308
+      "required_token_positions_per_second": 27836.719329101517,
+      "required_pages_per_second_upper_estimate": 0.0016951101591631452
     },
     {
       "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": 28541.724005420652,
-      "required_pages_per_second_upper_estimate": 0.002197220922486101
+      "required_token_positions_per_second": 27836.719329101517,
+      "required_pages_per_second_upper_estimate": 0.0021429477109252033
     },
     {
       "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": 28541.724005420652,
-      "required_pages_per_second_upper_estimate": 0.3543926891418932
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27836.719329101517,
+      "required_pages_per_second_upper_estimate": 0.34563889056087904
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "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": 688946,
-    "producer_batch_seconds": 0.18855529369528767,
-    "producer_p95_seconds": 0.2935596390016144,
-    "gpu_update_seconds": 0.9686871050518562,
-    "producer_batches_per_second": 5.303484088948661,
-    "gpu_batches_per_second": 1.032325087001615,
-    "producer_headroom": 5.13741664881226,
+    "producer_batch_seconds": 0.2596954910935665,
+    "producer_p95_seconds": 0.2751654009989579,
+    "gpu_update_seconds": 0.9932204895673815,
+    "producer_batches_per_second": 3.850663697660068,
+    "gpu_batches_per_second": 1.0068257859194703,
+    "producer_headroom": 3.8245580829492765,
     "sustainable": true,
-    "buffer_seconds": 60.05860051321508,
+    "buffer_seconds": 60.58644986361027,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 663227.07556234,
-    "upload_bytes_per_second": 5113595.432457817,
+    "generation_bytes_per_second": 613418.5249507198,
+    "upload_bytes_per_second": 5132690.98456202,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.15654563976598,
+    "estimated_full_shards_per_hour": 65.81266790099714,
     "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": 1.0242165819981892,
-      "bytes": 5354140,
-      "pages": 32,
-      "shards": 2
-    },
     {
       "seconds": 0.845649806999063,
       "bytes": 8068507,
@@ -562,6 +556,12 @@
       "bytes": 5835711,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.051623090999783,
+      "bytes": 7071478,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -228,3 +228,4 @@
 {"step": 62800, "loss": 0.6389019126072526, "lr": 0.0003, "grad_norm": 0.1355427224723376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9190803656, "tokens_per_second": 148913.4956377439, "data_wait_s": 1.2069998774677515e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 62900, "loss": 0.6282933112233877, "lr": 0.0003, "grad_norm": 0.12910738432293123, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9205508537, "tokens_per_second": 150436.70115853802, "data_wait_s": 8.184000762412325e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 63000, "loss": 0.6200405359268188, "lr": 0.0003, "grad_norm": 0.12977827007988665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9220130946, "tokens_per_second": 148037.63914714358, "data_wait_s": 1.4934001228539273e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 63100, "loss": 0.6327682584524155, "lr": 0.0003, "grad_norm": 0.12963534204895175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9234818345, "tokens_per_second": 150186.48587605206, "data_wait_s": 1.1703999916790053e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df3404e047e03da4892a5f4665e5238cda2561935612c6d2440bcf495953ad60
-size 490437
+oid sha256:27b9f9fe44e9f6c9e95b4df03ad59175751790dce5bdfdaf457961cf8c0153a9
+size 492555