Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77231.9780655203,
-      "required_pages_per_second_upper_estimate": 0.005343087445451145
+      "required_token_positions_per_second": 77441.39503454456,
+      "required_pages_per_second_upper_estimate": 0.005357575397282557
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9653.997258190038,
-      "required_pages_per_second_upper_estimate": 0.0040419270536169016
+      "required_token_positions_per_second": 9680.17437931807,
+      "required_pages_per_second_upper_estimate": 0.004052886867592758
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19307.994516380077,
-      "required_pages_per_second_upper_estimate": 0.004021346071705294
+      "required_token_positions_per_second": 19360.34875863614,
+      "required_pages_per_second_upper_estimate": 0.004032250079693017
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28961.991774570117,
-      "required_pages_per_second_upper_estimate": 0.0017362271844845987
+      "required_token_positions_per_second": 29040.523137954213,
+      "required_pages_per_second_upper_estimate": 0.0017409350198090265
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28961.991774570117,
-      "required_pages_per_second_upper_estimate": 0.0022516674701340365
+      "required_token_positions_per_second": 29040.523137954213,
+      "required_pages_per_second_upper_estimate": 0.0022577729382141877
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28961.991774570117,
-      "required_pages_per_second_upper_estimate": 0.3714933335202231
+      "required_token_positions_per_second": 29040.523137954213,
+      "required_pages_per_second_upper_estimate": 0.37250064952930584
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.19920899281251536,
-    "producer_p95_seconds": 0.21332920799977728,
-    "gpu_update_seconds": 0.954630476218702,
-    "producer_batches_per_second": 5.019853701791191,
-    "gpu_batches_per_second": 1.0475257441612456,
-    "producer_headroom": 4.7921053298891385,
+    "producer_batch_seconds": 0.120850244125009,
+    "producer_p95_seconds": 0.2179145519985468,
+    "gpu_update_seconds": 0.9520489651188732,
+    "producer_batches_per_second": 8.274704012724936,
+    "gpu_batches_per_second": 1.0503661435892004,
+    "producer_headroom": 7.8779233919797615,
     "sustainable": true,
-    "buffer_seconds": 60.141720001778225,
+    "buffer_seconds": 60.93113376760788,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 379934.0501074926,
-    "upload_bytes_per_second": 3523009.6668642773,
+    "generation_bytes_per_second": 380453.761562327,
+    "upload_bytes_per_second": 3516863.5123918983,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.52500274103721,
+    "estimated_full_shards_per_hour": 81.6365207209812,
     "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": 1.2970412000004217,
-      "bytes": 7490143,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 4.696426351999435,
       "bytes": 5652066,
@@ -562,6 +556,12 @@
       "bytes": 16491610,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.2731761789982556,
+      "bytes": 6636284,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -133,3 +133,4 @@
 {"step": 938300, "loss": 0.5379912257194519, "lr": 3e-05, "grad_norm": 0.3346158346146954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137340265616, "tokens_per_second": 153998.72302263198, "data_wait_s": 1.6156000128830783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 938400, "loss": 0.5361645817756653, "lr": 3e-05, "grad_norm": 0.3303286230450874, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137354912920, "tokens_per_second": 153196.51069038373, "data_wait_s": 1.4813000234426e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 938500, "loss": 0.5211808681488037, "lr": 3e-05, "grad_norm": 0.34861600819338995, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137369683862, "tokens_per_second": 154429.45637528106, "data_wait_s": 1.4294999346020631e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 938600, "loss": 0.48861345648765564, "lr": 3e-05, "grad_norm": 0.32073224374380005, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137384408236, "tokens_per_second": 154256.38199451554, "data_wait_s": 1.1725998774636537e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bcba94115fddf811f84b6a251272731d483e1c31399ccaec41e34f316d79ab05
-size 287661
+oid sha256:5123e9fe2a1005ff59ed175ebe8c36244fd5617d4ee8e6d64c15843c85032e71
+size 289779