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": 77151.93193928288,
-      "required_pages_per_second_upper_estimate": 0.005337549668705438
+      "required_token_positions_per_second": 76846.71237821915,
+      "required_pages_per_second_upper_estimate": 0.005316433871264096
     },
     {
       "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": 9643.99149241036,
-      "required_pages_per_second_upper_estimate": 0.004037737848429099
+      "required_token_positions_per_second": 9605.839047277394,
+      "required_pages_per_second_upper_estimate": 0.004021764216365578
     },
     {
       "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": 19287.98298482072,
-      "required_pages_per_second_upper_estimate": 0.004017178197421056
+      "required_token_positions_per_second": 19211.678094554787,
+      "required_pages_per_second_upper_estimate": 0.00400128590107394
     },
     {
       "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": 28931.974477231077,
-      "required_pages_per_second_upper_estimate": 0.0017344276933428826
+      "required_token_positions_per_second": 28817.51714183218,
+      "required_pages_per_second_upper_estimate": 0.0017275661508519505
     },
     {
       "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": 28931.974477231077,
-      "required_pages_per_second_upper_estimate": 0.0022493337572980633
+      "required_token_positions_per_second": 28817.51714183218,
+      "required_pages_per_second_upper_estimate": 0.0022404352029154117
     },
     {
       "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": 28931.974477231077,
-      "required_pages_per_second_upper_estimate": 0.37110830386002075
+      "required_token_positions_per_second": 28817.51714183218,
+      "required_pages_per_second_upper_estimate": 0.36964016805623556
     }
   ],
   "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": 0.11330320894563783,
-    "producer_p95_seconds": 0.32973902199591976,
-    "gpu_update_seconds": 0.955620917672192,
-    "producer_batches_per_second": 8.825875359627226,
-    "gpu_batches_per_second": 1.0464400490896657,
-    "producer_headroom": 8.434191110427356,
+    "producer_batch_seconds": 0.2968738611016306,
+    "producer_p95_seconds": 0.37027595799736446,
+    "gpu_update_seconds": 0.9594164502071387,
+    "producer_batches_per_second": 3.3684339749186067,
+    "gpu_batches_per_second": 1.0423002438452031,
+    "producer_headroom": 3.231730966973532,
     "sustainable": true,
-    "buffer_seconds": 60.2041178133481,
+    "buffer_seconds": 60.44323636304974,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472577.75103201496,
-    "upload_bytes_per_second": 5297202.370957243,
+    "generation_bytes_per_second": 446416.08439212065,
+    "upload_bytes_per_second": 5160096.919252535,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.40418432445847,
+    "estimated_full_shards_per_hour": 95.79049967596735,
     "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": 1.1696946029987885,
-      "bytes": 13219787,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.8621246889997565,
       "bytes": 1265428,
@@ -562,6 +556,12 @@
       "bytes": 1143669,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 2.9000736189991585,
+      "bytes": 10986050,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -379,3 +379,4 @@
 {"step": 435000, "loss": 0.5770679116249084, "lr": 0.0003, "grad_norm": 0.15449644206745677, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63671469675, "tokens_per_second": 152025.25322236057, "data_wait_s": 1.66549943969585e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 435100, "loss": 0.5555424690246582, "lr": 0.0003, "grad_norm": 0.16785926415375643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63686074299, "tokens_per_second": 142314.83712929187, "data_wait_s": 1.0607000149320811e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 435200, "loss": 0.6025435924530029, "lr": 0.0003, "grad_norm": 0.17389561632674724, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63700561761, "tokens_per_second": 149297.00401588806, "data_wait_s": 1.2103002518415451e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 435300, "loss": 0.5720965266227722, "lr": 0.0003, "grad_norm": 0.19244354152391166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 63715228629, "tokens_per_second": 151695.70800136589, "data_wait_s": 1.065700053004548e-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:f6b46bf84dbcf28ce510239d8151750eb6f764c800a54332e6b951c4b44ab313
-size 813387
+oid sha256:68252ad110e40b18e9d7d125bf42621fae69754d58431be4aa90c039a73f4097
+size 815505