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": 76901.79793107293,
-      "required_pages_per_second_upper_estimate": 0.005320244817626612
+      "required_token_positions_per_second": 76765.55632890557,
+      "required_pages_per_second_upper_estimate": 0.005310819307464605
     },
     {
       "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": 9612.724741384116,
-      "required_pages_per_second_upper_estimate": 0.004024647112698418
+      "required_token_positions_per_second": 9595.694541113196,
+      "required_pages_per_second_upper_estimate": 0.004017516923476006
     },
     {
       "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": 19225.449482768232,
-      "required_pages_per_second_upper_estimate": 0.0040041541180628705
+      "required_token_positions_per_second": 19191.389082226393,
+      "required_pages_per_second_upper_estimate": 0.003997060234863171
     },
     {
       "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": 28838.17422415235,
-      "required_pages_per_second_upper_estimate": 0.0017288045113955063
+      "required_token_positions_per_second": 28787.08362333959,
+      "required_pages_per_second_upper_estimate": 0.0017257417078876118
     },
     {
       "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": 28838.17422415235,
-      "required_pages_per_second_upper_estimate": 0.0022420412002049023
+      "required_token_positions_per_second": 28787.08362333959,
+      "required_pages_per_second_upper_estimate": 0.0022380691307155136
     },
     {
       "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": 28838.17422415235,
-      "required_pages_per_second_upper_estimate": 0.3699051349283918
+      "required_token_positions_per_second": 28787.08362333959,
+      "required_pages_per_second_upper_estimate": 0.3692497995579789
     }
   ],
   "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": 0.05265844196873104,
-    "producer_p95_seconds": 0.17604723399927025,
-    "gpu_update_seconds": 0.9587292102855955,
-    "producer_batches_per_second": 18.990307396367843,
-    "gpu_batches_per_second": 1.043047389473103,
-    "producer_headroom": 18.206562413200448,
+    "producer_batch_seconds": 0.06708011378908907,
+    "producer_p95_seconds": 0.1769912210002076,
+    "gpu_update_seconds": 0.9604307390688211,
+    "producer_batches_per_second": 14.907547759149079,
+    "gpu_batches_per_second": 1.0411994944784284,
+    "producer_headroom": 14.3176671120233,
     "sustainable": true,
-    "buffer_seconds": 60.39994024799252,
+    "buffer_seconds": 60.507136561335734,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 374382.28177509294,
-    "upload_bytes_per_second": 4961836.740156561,
+    "generation_bytes_per_second": 432985.9895226683,
+    "upload_bytes_per_second": 4974423.660555432,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.33372249545661,
+    "estimated_full_shards_per_hour": 92.90871395359075,
     "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": 0.9708695179999722,
-      "bytes": 43075,
-      "pages": 1,
-      "shards": 1
-    },
     {
       "seconds": 0.8249086619999844,
       "bytes": 105910,
@@ -562,6 +556,12 @@
       "bytes": 12583619,
       "pages": 38,
       "shards": 4
+    },
+    {
+      "seconds": 0.9289323200000581,
+      "bytes": 883918,
+      "pages": 25,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -76,3 +76,4 @@
 {"step": 404700, "loss": 0.635788083076477, "lr": 0.0003, "grad_norm": 0.1780795452564756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59237684063, "tokens_per_second": 153293.41920650937, "data_wait_s": 7.6200003604753874e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 404800, "loss": 0.624845027923584, "lr": 0.0003, "grad_norm": 0.16645235871608838, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59252363994, "tokens_per_second": 152852.3598016883, "data_wait_s": 1.3953000234323554e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 404900, "loss": 0.5621156692504883, "lr": 0.0003, "grad_norm": 0.1732311903236872, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59267149727, "tokens_per_second": 154028.61356223238, "data_wait_s": 7.122999704733957e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 405000, "loss": 0.5751948356628418, "lr": 0.0003, "grad_norm": 0.1561981491970473, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59281882394, "tokens_per_second": 153122.1400274555, "data_wait_s": 1.2495000191847794e-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 sha256:6ddbeecfe1c4b9b0ea136ec51a8e53e58eb4ca57117af26634c6d96aaf433475
-size 166152
+oid sha256:02940a6baab54e63129b63d7e5caada7459897c46fd3ad3df83488e149de17d5
+size 168270