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": 75572.1882942542,
-      "required_pages_per_second_upper_estimate": 0.0052282593378320886
+      "required_token_positions_per_second": 76005.15740222346,
+      "required_pages_per_second_upper_estimate": 0.0052582131453481435
     },
     {
       "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": 9446.523536781775,
-      "required_pages_per_second_upper_estimate": 0.003955062139007234
+      "required_token_positions_per_second": 9500.644675277932,
+      "required_pages_per_second_upper_estimate": 0.003977721529517687
     },
     {
       "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": 18893.04707356355,
-      "required_pages_per_second_upper_estimate": 0.0039349234622665565
+      "required_token_positions_per_second": 19001.289350555864,
+      "required_pages_per_second_upper_estimate": 0.003957467474023251
     },
     {
       "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": 28339.570610345327,
-      "required_pages_per_second_upper_estimate": 0.0016989139860714114
+      "required_token_positions_per_second": 28501.934025833798,
+      "required_pages_per_second_upper_estimate": 0.0017086474249153646
     },
     {
       "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": 28339.570610345327,
-      "required_pages_per_second_upper_estimate": 0.002203276962356927
+      "required_token_positions_per_second": 28501.934025833798,
+      "required_pages_per_second_upper_estimate": 0.0022159000037499654
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28339.570610345327,
-      "required_pages_per_second_upper_estimate": 0.36350958312932524
+      "required_token_positions_per_second": 28501.934025833798,
+      "required_pages_per_second_upper_estimate": 0.36559220669095827
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.40485083310943537,
-    "producer_p95_seconds": 0.3002608570022858,
-    "gpu_update_seconds": 0.9755969975743786,
-    "producer_batches_per_second": 2.4700455531227465,
-    "gpu_batches_per_second": 1.025013404598717,
-    "producer_headroom": 2.4097690254984965,
+    "producer_batch_seconds": 0.08032745317183299,
+    "producer_p95_seconds": 0.2951460379990749,
+    "gpu_update_seconds": 0.9700394357428587,
+    "producer_batches_per_second": 12.44904401314509,
+    "gpu_batches_per_second": 1.0308859239667896,
+    "producer_headroom": 12.076063630049276,
     "sustainable": true,
-    "buffer_seconds": 60.48701384961147,
+    "buffer_seconds": 60.14244501605724,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 294549.1564917664,
-    "upload_bytes_per_second": 3206208.7409059615,
+    "generation_bytes_per_second": 309025.57077535364,
+    "upload_bytes_per_second": 3237125.625473279,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.20339222969765,
+    "estimated_full_shards_per_hour": 66.30969374127824,
     "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": 1.7439238019978802,
-      "bytes": 6171251,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.5497890789993107,
       "bytes": 2026869,
@@ -562,6 +556,12 @@
       "bytes": 2439502,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.8466758049980854,
+      "bytes": 10680095,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -295,3 +295,4 @@
 {"step": 829500, "loss": 0.5509658455848694, "lr": 3e-05, "grad_norm": 0.2958763041286958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121415610041, "tokens_per_second": 151747.01386022512, "data_wait_s": 1.4760000340174884e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 829600, "loss": 0.5377776026725769, "lr": 3e-05, "grad_norm": 0.2788177400631082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121430194259, "tokens_per_second": 150181.48819472408, "data_wait_s": 1.1932999768760055e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 829700, "loss": 0.5087186694145203, "lr": 3e-05, "grad_norm": 0.3078371649183931, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121444782976, "tokens_per_second": 151170.97979522677, "data_wait_s": 1.2895998224848881e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 829800, "loss": 0.586624264717102, "lr": 3e-05, "grad_norm": 0.29802378772328575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 121459473518, "tokens_per_second": 152591.39854027424, "data_wait_s": 1.3722001312999055e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cd5e14de6192b5fbd4fc673ac22320191ec02fc36f3affbd82c8eb994a301286
-size 633126
+oid sha256:e9b7acb747150a1c4abc25bc5e9497ee6789c43c21d1cad1672dc59de01c0cb7
+size 635244