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": 76416.73429511857,
-      "required_pages_per_second_upper_estimate": 0.005286686989788591
+      "required_token_positions_per_second": 76903.33438556553,
+      "required_pages_per_second_upper_estimate": 0.005320351113113476
     },
     {
       "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": 9552.091786889821,
-      "required_pages_per_second_upper_estimate": 0.003999261360811727
+      "required_token_positions_per_second": 9612.91679819569,
+      "required_pages_per_second_upper_estimate": 0.004024727522874823
     },
     {
       "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": 19104.183573779643,
-      "required_pages_per_second_upper_estimate": 0.003978897627217623
+      "required_token_positions_per_second": 19225.83359639138,
+      "required_pages_per_second_upper_estimate": 0.004004234118800815
     },
     {
       "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": 28656.275360669464,
-      "required_pages_per_second_upper_estimate": 0.00171789995227319
+      "required_token_positions_per_second": 28838.750394587074,
+      "required_pages_per_second_upper_estimate": 0.0017288390519333066
     },
     {
       "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": 28656.275360669464,
-      "required_pages_per_second_upper_estimate": 0.0022278993636576524
+      "required_token_positions_per_second": 28838.750394587074,
+      "required_pages_per_second_upper_estimate": 0.002242085994921896
     },
     {
       "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": 28656.275360669464,
-      "required_pages_per_second_upper_estimate": 0.36757193161541624
+      "required_token_positions_per_second": 28838.750394587074,
+      "required_pages_per_second_upper_estimate": 0.36991252542408476
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.42164313707814927,
-    "producer_p95_seconds": 0.3130337810034689,
-    "gpu_update_seconds": 0.9648148495232107,
-    "producer_batches_per_second": 2.371673844686948,
-    "gpu_batches_per_second": 1.036468292848288,
-    "producer_headroom": 2.288226143579772,
+    "producer_batch_seconds": 0.1300550666172171,
+    "producer_p95_seconds": 0.3136580159989535,
+    "gpu_update_seconds": 0.9587100558001094,
+    "producer_batches_per_second": 7.689050692221296,
+    "gpu_batches_per_second": 1.0430682289708866,
+    "producer_headroom": 7.3715702181893485,
     "sustainable": true,
-    "buffer_seconds": 60.78333551996228,
+    "buffer_seconds": 60.39873351540689,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 362089.8009091521,
+    "generation_bytes_per_second": 362567.2373646163,
     "upload_bytes_per_second": 6182899.81192196,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.69604225593493,
+    "estimated_full_shards_per_hour": 77.79848900512569,
     "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": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -279,3 +279,4 @@
 {"step": 512900, "loss": 0.5696396827697754, "lr": 0.0003, "grad_norm": 0.17229861166229993, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75071428092, "tokens_per_second": 153775.07358882265, "data_wait_s": 2.3386997781926766e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 513000, "loss": 0.6181649565696716, "lr": 0.0003, "grad_norm": 0.1592788071664088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75086177840, "tokens_per_second": 137699.91849217133, "data_wait_s": 1.7480997485108674e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 513100, "loss": 0.62356036901474, "lr": 0.0003, "grad_norm": 0.17015670462856258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75100821852, "tokens_per_second": 152173.94342248706, "data_wait_s": 1.6312002117047086e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 513200, "loss": 0.6474027037620544, "lr": 0.0003, "grad_norm": 0.16722417052920815, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75115313829, "tokens_per_second": 150837.78690127746, "data_wait_s": 1.9545997929526493e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:658fc226141ef46179a9ede7c6935cf64eed993cdfb30e57eac971584a1c6347
-size 599238
+oid sha256:dbee61dad57e6645d378263e26e17dc20acef1ee81c2852c5d14be45d26074b1
+size 601356