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": 76729.09325194832,
-      "required_pages_per_second_upper_estimate": 0.005308296707194707
+      "required_token_positions_per_second": 76272.21105344521,
+      "required_pages_per_second_upper_estimate": 0.005276688536589506
     },
     {
       "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": 9591.13665649354,
-      "required_pages_per_second_upper_estimate": 0.00401560863236521
+      "required_token_positions_per_second": 9534.026381680651,
+      "required_pages_per_second_upper_estimate": 0.003991697752899207
     },
     {
       "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": 19182.27331298708,
-      "required_pages_per_second_upper_estimate": 0.003995161660529623
+      "required_token_positions_per_second": 19068.052763361302,
+      "required_pages_per_second_upper_estimate": 0.003971372532241029
     },
     {
       "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": 28773.409969480617,
-      "required_pages_per_second_upper_estimate": 0.001724921993217227
+      "required_token_positions_per_second": 28602.079145041953,
+      "required_pages_per_second_upper_estimate": 0.0017146509718991504
     },
     {
       "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": 28773.409969480617,
-      "required_pages_per_second_upper_estimate": 0.002237006064272026
+      "required_token_positions_per_second": 28602.079145041953,
+      "required_pages_per_second_upper_estimate": 0.002223685846276573
     },
     {
       "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": 28773.409969480617,
-      "required_pages_per_second_upper_estimate": 0.3690744086079016
+      "required_token_positions_per_second": 28602.079145041953,
+      "required_pages_per_second_upper_estimate": 0.3668767607527091
     }
   ],
   "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": 0.06902974213278412,
-    "producer_p95_seconds": 0.25711209700239124,
-    "gpu_update_seconds": 0.9608871534283104,
-    "producer_batches_per_second": 14.486509279962565,
-    "gpu_batches_per_second": 1.040704932345219,
-    "producer_headroom": 13.919900665136032,
+    "producer_batch_seconds": 0.12197898333593571,
+    "producer_p95_seconds": 0.28868629399948986,
+    "gpu_update_seconds": 0.9666430142996322,
+    "producer_batches_per_second": 8.198133585405891,
+    "gpu_batches_per_second": 1.03450807092889,
+    "producer_headroom": 7.924668560627802,
     "sustainable": true,
-    "buffer_seconds": 60.53589066598356,
+    "buffer_seconds": 60.898509900876824,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363494.8277837859,
-    "upload_bytes_per_second": 4941059.769888517,
+    "generation_bytes_per_second": 364567.9171694634,
+    "upload_bytes_per_second": 4908035.063720423,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.99752831587965,
+    "estimated_full_shards_per_hour": 78.22778831780364,
     "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": 1.5804328610010998,
-      "bytes": 12088466,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 0.7921914839989768,
       "bytes": 902002,
@@ -562,6 +556,12 @@
       "bytes": 5791149,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2756652919997578,
+      "bytes": 7797831,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -236,3 +236,4 @@
 {"step": 420700, "loss": 0.6293513774871826, "lr": 0.0003, "grad_norm": 0.17249792678666095, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61579348754, "tokens_per_second": 152410.75161214618, "data_wait_s": 1.5664998500142246e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420800, "loss": 0.5988439917564392, "lr": 0.0003, "grad_norm": 0.17409002378700117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61593903469, "tokens_per_second": 150870.86577655323, "data_wait_s": 1.058699854183942e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 420900, "loss": 0.5789002180099487, "lr": 0.0003, "grad_norm": 0.1462506781190776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61608463519, "tokens_per_second": 150293.09301309913, "data_wait_s": 1.3267999747768044e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 421000, "loss": 0.577822744846344, "lr": 0.0003, "grad_norm": 0.17032858961319572, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61623016538, "tokens_per_second": 150942.25230837657, "data_wait_s": 1.3732998922932893e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc87dc969403c79657b3ac3bde057ae7472028410b6b4179e892883da76f2fce
-size 508164
+oid sha256:db63e382fc40003d22380e436dc8cbbc18b93486ef26840de8e886d639aff73b
+size 510282