Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74993.06425637513,
-      "required_pages_per_second_upper_estimate": 0.006500208741558927
+      "required_token_positions_per_second": 75427.71143025016,
+      "required_pages_per_second_upper_estimate": 0.006537882856987223
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9374.133032046891,
-      "required_pages_per_second_upper_estimate": 0.0037394818222622034
+      "required_token_positions_per_second": 9428.46392878127,
+      "required_pages_per_second_upper_estimate": 0.0037611552292888422
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18748.266064093783,
-      "required_pages_per_second_upper_estimate": 0.004337081395714841
+      "required_token_positions_per_second": 18856.92785756254,
+      "required_pages_per_second_upper_estimate": 0.004362218389251589
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28122.399096140674,
-      "required_pages_per_second_upper_estimate": 0.0017125065581299291
+      "required_token_positions_per_second": 28285.39178634381,
+      "required_pages_per_second_upper_estimate": 0.0017224319578067447
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28122.399096140674,
-      "required_pages_per_second_upper_estimate": 0.002164940130204087
+      "required_token_positions_per_second": 28285.39178634381,
+      "required_pages_per_second_upper_estimate": 0.0021774877586885687
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28122.399096140674,
-      "required_pages_per_second_upper_estimate": 0.34918607715882977
+      "required_token_positions_per_second": 28285.39178634381,
+      "required_pages_per_second_upper_estimate": 0.3512099008697097
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.0828745698359512,
-    "producer_p95_seconds": 0.22042398700068588,
-    "gpu_update_seconds": 0.983130916586495,
-    "producer_batches_per_second": 12.066427638532337,
-    "gpu_batches_per_second": 1.0171585321231436,
-    "producer_headroom": 11.862878064194913,
+    "producer_batch_seconds": 0.06698707168749252,
+    "producer_p95_seconds": 0.21904598499895656,
+    "gpu_update_seconds": 0.9774656900226659,
+    "producer_batches_per_second": 14.928253688490683,
+    "gpu_batches_per_second": 1.0230538117167176,
+    "producer_headroom": 14.591855792453954,
     "sustainable": true,
-    "buffer_seconds": 60.954116828362686,
+    "buffer_seconds": 60.60287278140529,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340142.3925105288,
-    "upload_bytes_per_second": 5871483.152913433,
+    "generation_bytes_per_second": 340000.36231178755,
+    "upload_bytes_per_second": 5933834.7479872005,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.9866393231096,
+    "estimated_full_shards_per_hour": 72.95616294875354,
     "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": 0.8122863409998899,
-      "bytes": 1264426,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.096932616000231,
       "bytes": 5773373,
@@ -562,6 +556,12 @@
       "bytes": 7041796,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8336417800001072,
+      "bytes": 5838305,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -138,3 +138,4 @@
 {"step": 128800, "loss": 0.5997483246028423, "lr": 0.0003, "grad_norm": 0.1300345676672267, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18849397889, "tokens_per_second": 149107.47290031222, "data_wait_s": 1.2706999768852256e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 128900, "loss": 0.5837275832891464, "lr": 0.0003, "grad_norm": 0.1326213200135007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18863974238, "tokens_per_second": 149058.49035499312, "data_wait_s": 1.0752999514807016e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129000, "loss": 0.5921800285577774, "lr": 0.0003, "grad_norm": 0.14160159800155678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18878604005, "tokens_per_second": 149773.38357872202, "data_wait_s": 1.458100086892955e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 129100, "loss": 0.5809275228530169, "lr": 0.0003, "grad_norm": 0.1398279316633142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18893266134, "tokens_per_second": 135279.68346460993, "data_wait_s": 2.1052999727544375e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a0416435e1b2356373d6e29d622db3ac70fa96e78db0640eefeb5e4c7a27133a
-size 298251
+oid sha256:18f5d8b094ccb0e4e429c91aa439c0b3d2530a2edccca8c5128540460977088a
+size 300369