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": 77328.14030380835,
-      "required_pages_per_second_upper_estimate": 0.005349740172223046
+      "required_token_positions_per_second": 77172.91228226175,
+      "required_pages_per_second_upper_estimate": 0.005339001137513819
     },
     {
       "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": 9666.017537976044,
-      "required_pages_per_second_upper_estimate": 0.004046959693751316
+      "required_token_positions_per_second": 9646.614035282719,
+      "required_pages_per_second_upper_estimate": 0.004038835852364833
     },
     {
       "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": 19332.035075952088,
-      "required_pages_per_second_upper_estimate": 0.004026353086271954
+      "required_token_positions_per_second": 19293.228070565438,
+      "required_pages_per_second_upper_estimate": 0.004018270610459468
     },
     {
       "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": 28998.052613928132,
-      "required_pages_per_second_upper_estimate": 0.0017383889767423983
+      "required_token_positions_per_second": 28939.842105848158,
+      "required_pages_per_second_upper_estimate": 0.0017348993456653033
     },
     {
       "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": 28998.052613928132,
-      "required_pages_per_second_upper_estimate": 0.0022544710417792524
+      "required_token_positions_per_second": 28939.842105848158,
+      "required_pages_per_second_upper_estimate": 0.0022499454308170003
     },
     {
       "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": 28998.052613928132,
-      "required_pages_per_second_upper_estimate": 0.3719558832483951
+      "required_token_positions_per_second": 28939.842105848158,
+      "required_pages_per_second_upper_estimate": 0.3712092213523192
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05433913745318364,
-    "producer_p95_seconds": 0.1885013289993367,
-    "gpu_update_seconds": 0.9534433352507373,
-    "producer_batches_per_second": 18.402942094205283,
-    "gpu_batches_per_second": 1.0488300279921923,
-    "producer_headroom": 17.546162488725273,
+    "producer_batch_seconds": 0.1038673810625994,
+    "producer_p95_seconds": 0.2178431440006534,
+    "gpu_update_seconds": 0.955361121144918,
+    "producer_batches_per_second": 9.627661637076553,
+    "gpu_batches_per_second": 1.04672461320342,
+    "producer_headroom": 9.197893615601373,
     "sustainable": true,
-    "buffer_seconds": 60.06693012079645,
+    "buffer_seconds": 60.187750632129834,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 378462.8794478001,
-    "upload_bytes_per_second": 3494888.9041931555,
+    "generation_bytes_per_second": 443568.08380255336,
+    "upload_bytes_per_second": 3519811.0172344637,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.2093237645674,
+    "estimated_full_shards_per_hour": 95.17938504750681,
     "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": [
@@ -526,6 +526,12 @@
       "bytes": 5988822,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0547332340011053,
+      "bytes": 6465464,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -104,3 +104,4 @@
 {"step": 655400, "loss": 0.5680849552154541, "lr": 0.0001, "grad_norm": 0.18475815235608337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95933072724, "tokens_per_second": 153350.9804860552, "data_wait_s": 1.1686001016641967e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 655500, "loss": 0.5729182958602905, "lr": 0.0001, "grad_norm": 0.18111353609334457, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95947683842, "tokens_per_second": 152825.6094391479, "data_wait_s": 1.31070009956602e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 655600, "loss": 0.5763890147209167, "lr": 0.0001, "grad_norm": 0.17778064182595044, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95962405390, "tokens_per_second": 153297.05813810648, "data_wait_s": 1.7561000277055427e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 655700, "loss": 0.481076180934906, "lr": 0.0001, "grad_norm": 0.18059763053049702, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95977171138, "tokens_per_second": 74571.72193907037, "data_wait_s": 1.0036999810836278e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:509f58ef48c0de605cd9e9a72753a2e645891535374c1b918c02c8b8cc469ca3
-size 226239
+oid sha256:2eeb174bb2f795ca1f8633f88a214609c757adb0cb2b15d04b6b7974efa5e9ee
+size 228357