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": 76173.77610189145,
-      "required_pages_per_second_upper_estimate": 0.005269878578240462
+      "required_token_positions_per_second": 76020.22415053255,
+      "required_pages_per_second_upper_estimate": 0.005259255497955839
     },
     {
       "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": 9521.72201273643,
-      "required_pages_per_second_upper_estimate": 0.003986546170566704
+      "required_token_positions_per_second": 9502.528018816569,
+      "required_pages_per_second_upper_estimate": 0.003978510046128629
     },
     {
       "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": 19043.44402547286,
-      "required_pages_per_second_upper_estimate": 0.003966247181114927
+      "required_token_positions_per_second": 19005.056037633138,
+      "required_pages_per_second_upper_estimate": 0.00395825197560723
     },
     {
       "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": 28565.166038209292,
-      "required_pages_per_second_upper_estimate": 0.001712438087507583
+      "required_token_positions_per_second": 28507.58405644971,
+      "required_pages_per_second_upper_estimate": 0.0017089861356236913
     },
     {
       "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": 28565.166038209292,
-      "required_pages_per_second_upper_estimate": 0.0022208160145839313
+      "required_token_positions_per_second": 28507.58405644971,
+      "required_pages_per_second_upper_estimate": 0.0022163392687784946
     },
     {
       "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": 28565.166038209292,
-      "required_pages_per_second_upper_estimate": 0.3664032790524659
+      "required_token_positions_per_second": 28507.58405644971,
+      "required_pages_per_second_upper_estimate": 0.36566467921716894
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09934504785155696,
-    "producer_p95_seconds": 0.24446639199959463,
-    "gpu_update_seconds": 0.9678921509861881,
-    "producer_batches_per_second": 10.065927005181143,
-    "gpu_batches_per_second": 1.0331729614514356,
-    "producer_headroom": 9.742731740714735,
+    "producer_batch_seconds": 0.08479652061703291,
+    "producer_p95_seconds": 0.22563351900316775,
+    "gpu_update_seconds": 0.9698471797979236,
+    "producer_batches_per_second": 11.792936699800533,
+    "gpu_batches_per_second": 1.0310902798195063,
+    "producer_headroom": 11.43734639983698,
     "sustainable": true,
-    "buffer_seconds": 60.00931336114366,
+    "buffer_seconds": 60.13052514747126,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 370484.75028339017,
-    "upload_bytes_per_second": 5591680.691984401,
+    "generation_bytes_per_second": 370991.44447073084,
+    "upload_bytes_per_second": 5615990.035108827,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.49740296722678,
+    "estimated_full_shards_per_hour": 79.60612774459308,
     "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": 1.0064647640010662,
-      "bytes": 5698061,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.1021013760000642,
       "bytes": 5987765,
@@ -562,6 +556,12 @@
       "bytes": 11819263,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.0389532609988237,
+      "bytes": 7773400,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -179,3 +179,4 @@
 {"step": 502900, "loss": 0.6077935695648193, "lr": 0.0003, "grad_norm": 0.15652726329904404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73607510736, "tokens_per_second": 150449.8674342424, "data_wait_s": 1.1202002497157082e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 503000, "loss": 0.6127749085426331, "lr": 0.0003, "grad_norm": 0.15825898452385917, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73622088908, "tokens_per_second": 151817.96304782038, "data_wait_s": 8.316997991641983e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 503100, "loss": 0.6267172694206238, "lr": 0.0003, "grad_norm": 0.17689502830195086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73636823421, "tokens_per_second": 153319.1022309895, "data_wait_s": 1.2383003195282072e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 503200, "loss": 0.5751982927322388, "lr": 0.0003, "grad_norm": 0.15358865132269717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 73651566299, "tokens_per_second": 145755.68371992742, "data_wait_s": 1.2100997992092744e-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:af5c9cb22b6ba0e9f8e1bf9b4a773fc4851e55fcf7a31334a0d3d89ad09abeb4
-size 385872
+oid sha256:640aa4e16e56dfffa924b0dc467dceb96d48943ec8d6af981206f02f4970e67c
+size 387990