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": 73863.57629496043,
-      "required_pages_per_second_upper_estimate": 0.006402307587724566
+      "required_token_positions_per_second": 74022.93227515588,
+      "required_pages_per_second_upper_estimate": 0.006416120160203868
     },
     {
       "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": 9232.947036870053,
-      "required_pages_per_second_upper_estimate": 0.0036831606178674217
+      "required_token_positions_per_second": 9252.866534394485,
+      "required_pages_per_second_upper_estimate": 0.003691106803252946
     },
     {
       "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": 18465.894073740106,
-      "required_pages_per_second_upper_estimate": 0.004271759605323816
+      "required_token_positions_per_second": 18505.73306878897,
+      "required_pages_per_second_upper_estimate": 0.004280975655686005
     },
     {
       "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": 27698.84111061016,
-      "required_pages_per_second_upper_estimate": 0.0016867140990481273
+      "required_token_positions_per_second": 27758.599603183455,
+      "required_pages_per_second_upper_estimate": 0.0016903530777172611
     },
     {
       "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": 27698.84111061016,
-      "required_pages_per_second_upper_estimate": 0.0021323334639944016
+      "required_token_positions_per_second": 27758.599603183455,
+      "required_pages_per_second_upper_estimate": 0.002136933838174789
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27698.84111061016,
-      "required_pages_per_second_upper_estimate": 0.34392690453592956
+      "required_token_positions_per_second": 27758.599603183455,
+      "required_pages_per_second_upper_estimate": 0.3446689050148808
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.11786020207830461,
-    "producer_p95_seconds": 0.29016289100036374,
-    "gpu_update_seconds": 0.9981645040524572,
-    "producer_batches_per_second": 8.48462824911512,
-    "gpu_batches_per_second": 1.0018388711881567,
-    "producer_headroom": 8.469054748347462,
+    "producer_batch_seconds": 0.10038593777355231,
+    "producer_p95_seconds": 0.25421538400041754,
+    "gpu_update_seconds": 0.9960156634425186,
+    "producer_batches_per_second": 9.961554597973384,
+    "gpu_batches_per_second": 1.0040002749994015,
+    "producer_headroom": 9.92186441181933,
     "sustainable": true,
-    "buffer_seconds": 60.88803474719989,
+    "buffer_seconds": 60.756955469993635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 418892.7533661353,
-    "upload_bytes_per_second": 5731055.085245443,
+    "generation_bytes_per_second": 433625.8004065467,
+    "upload_bytes_per_second": 5656184.122481296,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.88463354814571,
+    "estimated_full_shards_per_hour": 93.04600247523594,
     "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": 1.0918037890005508,
-      "bytes": 11651730,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.0898228620008013,
       "bytes": 2320314,
@@ -562,6 +556,12 @@
       "bytes": 12412930,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0742992420018709,
+      "bytes": 6000047,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -198,3 +198,4 @@
 {"step": 334800, "loss": 0.5978746525943279, "lr": 0.0003, "grad_norm": 0.16581356789209187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49006190754, "tokens_per_second": 148435.20168036572, "data_wait_s": 9.647999831940979e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 334900, "loss": 0.6604721676558256, "lr": 0.0003, "grad_norm": 0.1942763048124513, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49020784092, "tokens_per_second": 147597.65132997284, "data_wait_s": 1.2881999282399192e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 335000, "loss": 0.6121786795556545, "lr": 0.0003, "grad_norm": 0.162747351060619, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49035397952, "tokens_per_second": 147883.44878684354, "data_wait_s": 7.835002179490402e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 335100, "loss": 0.5736087337136269, "lr": 0.0003, "grad_norm": 0.17493034646593886, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 49050095421, "tokens_per_second": 131968.2329283222, "data_wait_s": 1.3155000488040969e-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 sha256:0b47cc7196231f1c7f664a8feee87737762582def1bb5af71f732e4b19c62fe0
-size 426897
+oid sha256:26520f467171b6b97fc903273bb2525c1ba74ccb409dd57eb2ff19d1eb2a5557
+size 429015