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": 76231.95896826015,
-      "required_pages_per_second_upper_estimate": 0.0052739038039386985
+      "required_token_positions_per_second": 76666.73379054533,
+      "required_pages_per_second_upper_estimate": 0.005303982534960962
     },
     {
       "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": 9528.994871032519,
-      "required_pages_per_second_upper_estimate": 0.003989591164460724
+      "required_token_positions_per_second": 9583.341723818166,
+      "required_pages_per_second_upper_estimate": 0.004012345056830736
     },
     {
       "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": 19057.989742065038,
-      "required_pages_per_second_upper_estimate": 0.003969276670284735
+      "required_token_positions_per_second": 19166.683447636333,
+      "required_pages_per_second_upper_estimate": 0.003991914702709459
     },
     {
       "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": 28586.984613097557,
-      "required_pages_per_second_upper_estimate": 0.0017137460777570995
+      "required_token_positions_per_second": 28750.0251714545,
+      "required_pages_per_second_upper_estimate": 0.0017235201102820804
     },
     {
       "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": 28586.984613097557,
-      "required_pages_per_second_upper_estimate": 0.0022225123128117256
+      "required_token_positions_per_second": 28750.0251714545,
+      "required_pages_per_second_upper_estimate": 0.0022351879990843513
     },
     {
       "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": 28586.984613097557,
-      "required_pages_per_second_upper_estimate": 0.36668314430417204
+      "required_token_positions_per_second": 28750.0251714545,
+      "required_pages_per_second_upper_estimate": 0.36877445352746246
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08202498109365308,
-    "producer_p95_seconds": 0.1987917610003933,
-    "gpu_update_seconds": 0.9671534222372182,
-    "producer_batches_per_second": 12.191407869490845,
-    "gpu_batches_per_second": 1.0339621170825215,
-    "producer_headroom": 11.790961842867825,
+    "producer_batch_seconds": 0.11883315531260052,
+    "producer_p95_seconds": 0.19728369300173654,
+    "gpu_update_seconds": 0.9616687232486778,
+    "producer_batches_per_second": 8.415159871581434,
+    "gpu_batches_per_second": 1.0398591280184644,
+    "producer_headroom": 8.092596049637224,
     "sustainable": true,
-    "buffer_seconds": 60.93066560094475,
+    "buffer_seconds": 60.5851295646667,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 243152.63476515617,
-    "upload_bytes_per_second": 4975445.243252082,
+    "generation_bytes_per_second": 243506.71165335423,
+    "upload_bytes_per_second": 5014732.360714136,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.17489511695875,
+    "estimated_full_shards_per_hour": 52.25087177467795,
     "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": 0.8767920650002452,
-      "bytes": 2456869,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.450166357999933,
       "bytes": 12214165,
@@ -562,6 +556,12 @@
       "bytes": 1063767,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.2301429309991363,
+      "bytes": 7644122,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -103,3 +103,4 @@
 {"step": 450500, "loss": 0.6055036187171936, "lr": 0.0003, "grad_norm": 0.1672120317372175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65938946967, "tokens_per_second": 150269.87771515214, "data_wait_s": 1.5751000319141895e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 450600, "loss": 0.5948117971420288, "lr": 0.0003, "grad_norm": 0.14550870833572543, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65953663832, "tokens_per_second": 153184.70147082658, "data_wait_s": 1.693900048849173e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 450700, "loss": 0.6282864212989807, "lr": 0.0003, "grad_norm": 0.16398398248758808, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65968400941, "tokens_per_second": 153336.29620067237, "data_wait_s": 1.0555999324424192e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 450800, "loss": 0.6025879383087158, "lr": 0.0003, "grad_norm": 0.15839335428853707, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 65983024556, "tokens_per_second": 152071.65040896164, "data_wait_s": 1.1902000551344827e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:934884919c54c3dc708d8813badc583ea193b8e0959cee4cb7c86e2396682fa5
-size 224121
+oid sha256:1bb51d79f76027bc0f79c7fc202af0d891b055060422b8d2c0c81e02c03e8722
+size 226239