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": 76057.26495648404,
-      "required_pages_per_second_upper_estimate": 0.0052618180668580735
+      "required_token_positions_per_second": 75887.92824925356,
+      "required_pages_per_second_upper_estimate": 0.005250102960536575
     },
     {
       "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": 9507.158119560505,
-      "required_pages_per_second_upper_estimate": 0.003980448572622616
+      "required_token_positions_per_second": 9485.991031156695,
+      "required_pages_per_second_upper_estimate": 0.003971586354727011
     },
     {
       "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": 19014.31623912101,
-      "required_pages_per_second_upper_estimate": 0.003960180631369218
+      "required_token_positions_per_second": 18971.98206231339,
+      "required_pages_per_second_upper_estimate": 0.003951363538767509
     },
     {
       "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": 28521.474358681517,
-      "required_pages_per_second_upper_estimate": 0.0017098188380332247
+      "required_token_positions_per_second": 28457.973093470086,
+      "required_pages_per_second_upper_estimate": 0.001706012034144617
     },
     {
       "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": 28521.474358681517,
-      "required_pages_per_second_upper_estimate": 0.0022174191786800347
+      "required_token_positions_per_second": 28457.973093470086,
+      "required_pages_per_second_upper_estimate": 0.002212482234622393
     },
     {
       "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": 28521.474358681517,
-      "required_pages_per_second_upper_estimate": 0.365842849099954
+      "required_token_positions_per_second": 28457.973093470086,
+      "required_pages_per_second_upper_estimate": 0.3650283230521682
     }
   ],
   "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": 0.07555186401569358,
-    "producer_p95_seconds": 0.2600243739980215,
-    "gpu_update_seconds": 0.9693748525164989,
-    "producer_batches_per_second": 13.235940807393987,
-    "gpu_batches_per_second": 1.0315926779037006,
-    "producer_headroom": 12.830588168084656,
+    "producer_batch_seconds": 0.11742779875027054,
+    "producer_p95_seconds": 0.2636444810013927,
+    "gpu_update_seconds": 0.9715379204692571,
+    "producer_batches_per_second": 8.515871119466897,
+    "gpu_batches_per_second": 1.0292959018182177,
+    "producer_headroom": 8.273491718391075,
     "sustainable": true,
-    "buffer_seconds": 60.10124085602293,
+    "buffer_seconds": 60.235351069093944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301787.7805973603,
-    "upload_bytes_per_second": 5190487.162104371,
+    "generation_bytes_per_second": 331872.2560295619,
+    "upload_bytes_per_second": 5139200.681578755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.75663245621305,
+    "estimated_full_shards_per_hour": 71.2120605532183,
     "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": 1.4776745150011266,
-      "bytes": 12319008,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.1878520110003592,
       "bytes": 6586003,
@@ -562,6 +556,12 @@
       "bytes": 6107817,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.4312592239984951,
+      "bytes": 7807529,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -211,3 +211,4 @@
 {"step": 625500, "loss": 0.5510759353637695, "lr": 0.0001, "grad_norm": 0.16776120099390332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91556609596, "tokens_per_second": 151143.33589475485, "data_wait_s": 1.6912999853957444e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625600, "loss": 0.5008061528205872, "lr": 0.0001, "grad_norm": 0.16755687114194404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91571384588, "tokens_per_second": 153110.7857187314, "data_wait_s": 1.8451002688379958e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 625700, "loss": 0.529500424861908, "lr": 0.0001, "grad_norm": 0.1569622526422912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91586040695, "tokens_per_second": 151100.42773856962, "data_wait_s": 1.3044002116657794e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 625800, "loss": 0.5401984453201294, "lr": 0.0001, "grad_norm": 0.16631244845804258, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91600588830, "tokens_per_second": 150900.90455086165, "data_wait_s": 1.777400029823184e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e3db28e693e342a304035dcc843d1403b498958a987c7ca42f848a42baecb3cf
-size 455214
+oid sha256:cb2febf2e8d6b7a0da307c21b8b8dbc83fcc7cbbf1ab6306b169fb45587fa438
+size 457332