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": 77393.57426711687,
-      "required_pages_per_second_upper_estimate": 0.005354267045632442
+      "required_token_positions_per_second": 77021.01807998876,
+      "required_pages_per_second_upper_estimate": 0.005328492744157467
     },
     {
       "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": 9674.196783389609,
-      "required_pages_per_second_upper_estimate": 0.004050384173003909
+      "required_token_positions_per_second": 9627.627259998595,
+      "required_pages_per_second_upper_estimate": 0.0040308864860423245
     },
     {
       "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": 19348.393566779218,
-      "required_pages_per_second_upper_estimate": 0.0040297601285088315
+      "required_token_positions_per_second": 19255.25451999719,
+      "required_pages_per_second_upper_estimate": 0.004010361721305974
     },
     {
       "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": 29022.590350168826,
-      "required_pages_per_second_upper_estimate": 0.0017398599765630742
+      "required_token_positions_per_second": 28882.881779995787,
+      "required_pages_per_second_upper_estimate": 0.0017314846611038352
     },
     {
       "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": 29022.590350168826,
-      "required_pages_per_second_upper_estimate": 0.002256378742841871
+      "required_token_positions_per_second": 28882.881779995787,
+      "required_pages_per_second_upper_estimate": 0.002245517016023974
     },
     {
       "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": 29022.590350168826,
-      "required_pages_per_second_upper_estimate": 0.3722706269823223
+      "required_token_positions_per_second": 28882.881779995787,
+      "required_pages_per_second_upper_estimate": 0.3704785954515179
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05365187114843195,
-    "producer_p95_seconds": 0.18569849999948929,
-    "gpu_update_seconds": 0.9526372272914354,
-    "producer_batches_per_second": 18.638678923861285,
-    "gpu_batches_per_second": 1.0497175329198793,
-    "producer_headroom": 17.75589941040253,
+    "producer_batch_seconds": 0.0992656925156723,
+    "producer_p95_seconds": 0.19791945299948566,
+    "gpu_update_seconds": 0.9572452018672506,
+    "producer_batches_per_second": 10.073973944644749,
+    "gpu_batches_per_second": 1.0446644162324865,
+    "producer_headroom": 9.643263222246885,
     "sustainable": true,
-    "buffer_seconds": 60.01614531936043,
+    "buffer_seconds": 60.306447717636786,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 350743.5815092422,
-    "upload_bytes_per_second": 4807615.500962383,
+    "generation_bytes_per_second": 314308.2585646183,
+    "upload_bytes_per_second": 4770445.77834912,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.26140769918393,
+    "estimated_full_shards_per_hour": 67.44323556617653,
     "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": 1.7505123790000425,
-      "bytes": 15799691,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 1.5947456709999415,
       "bytes": 2710887,
@@ -562,6 +556,12 @@
       "bytes": 1163402,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.532974641999317,
+      "bytes": 11431908,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -82,3 +82,4 @@
 {"step": 850700, "loss": 0.506303071975708, "lr": 3e-05, "grad_norm": 0.2909025878885348, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124519048940, "tokens_per_second": 151697.43609105863, "data_wait_s": 9.502000466454774e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 850800, "loss": 0.6000729203224182, "lr": 3e-05, "grad_norm": 0.30744469547091996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124533588924, "tokens_per_second": 151344.06133344714, "data_wait_s": 8.895000064512715e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 850900, "loss": 0.5280565619468689, "lr": 3e-05, "grad_norm": 0.2985700083009807, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124548224086, "tokens_per_second": 151711.9943240079, "data_wait_s": 1.1285001164651476e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 851000, "loss": 0.5287325978279114, "lr": 3e-05, "grad_norm": 0.29213773795163894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 124562820078, "tokens_per_second": 151988.8861631096, "data_wait_s": 1.1604999599512666e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8da18fb3e5cfb9002fb2227820f1be12b04afabbb43c5e1707f244e8f8c3615e
-size 179643
+oid sha256:879ac4b31e8533cc5a77b93575788bc751b8560cc3e870c207d92d35b45b8dd6
+size 181761