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": 76110.80031888915,
-      "required_pages_per_second_upper_estimate": 0.005265521767448414
+      "required_token_positions_per_second": 76612.7067856645,
+      "required_pages_per_second_upper_estimate": 0.0053002448214033845
     },
     {
       "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": 9513.850039861143,
-      "required_pages_per_second_upper_estimate": 0.003983250339909307
+      "required_token_positions_per_second": 9576.588348208063,
+      "required_pages_per_second_upper_estimate": 0.0040095175594893045
     },
     {
       "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": 19027.700079722286,
-      "required_pages_per_second_upper_estimate": 0.003962968132410852
+      "required_token_positions_per_second": 19153.176696416125,
+      "required_pages_per_second_upper_estimate": 0.003989101602627273
     },
     {
       "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": 28541.55011958343,
-      "required_pages_per_second_upper_estimate": 0.0017110223492453824
+      "required_token_positions_per_second": 28729.765044624186,
+      "required_pages_per_second_upper_estimate": 0.0017223055466140255
     },
     {
       "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": 28541.55011958343,
-      "required_pages_per_second_upper_estimate": 0.0022189799807862186
+      "required_token_positions_per_second": 28729.765044624186,
+      "required_pages_per_second_upper_estimate": 0.0022336128633381742
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28541.55011958343,
-      "required_pages_per_second_upper_estimate": 0.36610035940513114
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28729.765044624186,
+      "required_pages_per_second_upper_estimate": 0.3685145783741125
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09076385435159295,
-    "producer_p95_seconds": 0.22895767299996805,
-    "gpu_update_seconds": 0.9686930066573248,
-    "producer_batches_per_second": 11.017601744041068,
-    "gpu_batches_per_second": 1.0323187977279886,
-    "producer_headroom": 10.672673759588127,
+    "producer_batch_seconds": 0.06635057974222036,
+    "producer_p95_seconds": 0.2209095759972115,
+    "gpu_update_seconds": 0.9623468885685648,
+    "producer_batches_per_second": 15.071458363817092,
+    "gpu_batches_per_second": 1.0391263398663262,
+    "producer_headroom": 14.50397106261005,
     "sustainable": true,
-    "buffer_seconds": 60.05896641275414,
+    "buffer_seconds": 60.62785397981958,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 268899.7435005941,
-    "upload_bytes_per_second": 4805712.845759498,
+    "generation_bytes_per_second": 286857.3535846135,
+    "upload_bytes_per_second": 4837786.797908241,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 57.699625289567635,
+    "estimated_full_shards_per_hour": 61.552910381830245,
     "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": 1.1272584719990846,
-      "bytes": 2509117,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.3743308060002164,
       "bytes": 6082381,
@@ -562,6 +556,12 @@
       "bytes": 8271893,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 1.2550481799989939,
+      "bytes": 6009424,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -200,3 +200,4 @@
 {"step": 945000, "loss": 0.547132670879364, "lr": 3e-05, "grad_norm": 0.3489687738519755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138321369205, "tokens_per_second": 150205.0161449377, "data_wait_s": 1.516900010756217e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945100, "loss": 0.5197945237159729, "lr": 3e-05, "grad_norm": 0.34259761542248635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138336025394, "tokens_per_second": 141410.8378124374, "data_wait_s": 1.3495999155566096e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 945200, "loss": 0.5494606494903564, "lr": 3e-05, "grad_norm": 0.3465691098478906, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138350543162, "tokens_per_second": 151311.5287856669, "data_wait_s": 1.5022000297904015e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 945300, "loss": 0.5615050196647644, "lr": 3e-05, "grad_norm": 0.3714581616074633, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138365298444, "tokens_per_second": 154376.5902163234, "data_wait_s": 1.456500103813596e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:972e2e92dad5053da2b493a344a04ad0bb2ad9501f64c73aa2030ce7d99c765a
-size 431133
+oid sha256:283e42a5068e114d3a59452bce8883bcc83f52b676411183240b7b4cee3ccdc4
+size 433251