Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75217.60175351969,
-      "required_pages_per_second_upper_estimate": 0.006519671082726333
+      "required_token_positions_per_second": 75105.56417247857,
+      "required_pages_per_second_upper_estimate": 0.006509959949158341
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9402.200219189961,
-      "required_pages_per_second_upper_estimate": 0.0037506782428554175
+      "required_token_positions_per_second": 9388.195521559821,
+      "required_pages_per_second_upper_estimate": 0.0037450915595818655
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18804.400438379922,
-      "required_pages_per_second_upper_estimate": 0.0043500670952746935
+      "required_token_positions_per_second": 18776.391043119642,
+      "required_pages_per_second_upper_estimate": 0.0043435876146297485
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28206.60065756988,
-      "required_pages_per_second_upper_estimate": 0.0017176339914495168
+      "required_token_positions_per_second": 28164.58656467946,
+      "required_pages_per_second_upper_estimate": 0.0017150755536233997
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28206.60065756988,
-      "required_pages_per_second_upper_estimate": 0.0021714221994877476
+      "required_token_positions_per_second": 28164.58656467946,
+      "required_pages_per_second_upper_estimate": 0.00216818783831459
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28206.60065756988,
-      "required_pages_per_second_upper_estimate": 0.3502315787472824
+      "required_token_positions_per_second": 28164.58656467946,
+      "required_pages_per_second_upper_estimate": 0.3497099043257069
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.062151386586023705,
-    "producer_p95_seconds": 0.2066321929996775,
-    "gpu_update_seconds": 0.9801961014603875,
-    "producer_batches_per_second": 16.089745618400652,
-    "gpu_batches_per_second": 1.0202040168391884,
-    "producer_headroom": 15.771105928645671,
+    "producer_batch_seconds": 0.08027491864842773,
+    "producer_p95_seconds": 0.2388077310006338,
+    "gpu_update_seconds": 0.9816582940604106,
+    "producer_batches_per_second": 12.457191073336405,
+    "gpu_batches_per_second": 1.0186844098914736,
+    "producer_headroom": 12.22870493783599,
     "sustainable": true,
-    "buffer_seconds": 60.77215829054403,
+    "buffer_seconds": 60.86281423174546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 346105.4722208976,
-    "upload_bytes_per_second": 4117364.844515269,
+    "generation_bytes_per_second": 346049.8258762993,
+    "upload_bytes_per_second": 4075090.1543999133,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.26617741556355,
+    "estimated_full_shards_per_hour": 74.25423700539335,
     "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": 1.0240621299999475,
-      "bytes": 6061414,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.345420446999924,
       "bytes": 12056021,
@@ -562,6 +556,12 @@
       "bytes": 7818449,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 0.8705007850003312,
+      "bytes": 1070987,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -141,3 +141,4 @@
 {"step": 329100, "loss": 0.5838132090866566, "lr": 0.0003, "grad_norm": 0.15181576133042385, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48171621084, "tokens_per_second": 148798.20042659793, "data_wait_s": 1.4307999663287774e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329200, "loss": 0.6187696345150471, "lr": 0.0003, "grad_norm": 0.20134963701953068, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48186371470, "tokens_per_second": 150530.256790889, "data_wait_s": 2.0486000721575692e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 329300, "loss": 0.5998435486108065, "lr": 0.0003, "grad_norm": 0.14355735746012452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48201268512, "tokens_per_second": 151673.6051440547, "data_wait_s": 1.831900044635404e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 329400, "loss": 0.6524562239646912, "lr": 0.0003, "grad_norm": 0.15385772274837856, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 48215914009, "tokens_per_second": 149490.48071541806, "data_wait_s": 1.3350001609069295e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:640de12277f53308d1bca9e0c6f1f6d2b04a45f9c29364a5dbe8e9a50eedadd0
-size 304605
+oid sha256:5744d326c1408819660025dfd78c58c204cc5722f1dbac897ae74c6aed4fbd0a
+size 306723