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": 76058.16431604026,
-      "required_pages_per_second_upper_estimate": 0.006592528915259278
+      "required_token_positions_per_second": 75941.1267787343,
+      "required_pages_per_second_upper_estimate": 0.006582384398154523
     },
     {
       "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": 9507.270539505033,
-      "required_pages_per_second_upper_estimate": 0.0037925923645703817
+      "required_token_positions_per_second": 9492.640847341787,
+      "required_pages_per_second_upper_estimate": 0.003786756361633073
     },
     {
       "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": 19014.541079010065,
-      "required_pages_per_second_upper_estimate": 0.0043986794341355166
+      "required_token_positions_per_second": 18985.281694683574,
+      "required_pages_per_second_upper_estimate": 0.004391910790519159
     },
     {
       "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": 28521.811618515098,
-      "required_pages_per_second_upper_estimate": 0.0017368286851869796
+      "required_token_positions_per_second": 28477.92254202536,
+      "required_pages_per_second_upper_estimate": 0.001734156070696943
     },
     {
       "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": 28521.811618515098,
-      "required_pages_per_second_upper_estimate": 0.0021956880118210965
+      "required_token_positions_per_second": 28477.92254202536,
+      "required_pages_per_second_upper_estimate": 0.0021923093092201767
     },
     {
       "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": 28521.811618515098,
-      "required_pages_per_second_upper_estimate": 0.3541454439389982
+      "required_token_positions_per_second": 28477.92254202536,
+      "required_pages_per_second_upper_estimate": 0.35360048849628567
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.05721442383584474,
-    "producer_p95_seconds": 0.20357603699994797,
-    "gpu_update_seconds": 0.9693633900187512,
-    "producer_batches_per_second": 17.47811011553876,
-    "gpu_batches_per_second": 1.031604876248376,
-    "producer_headroom": 16.94264007271968,
+    "producer_batch_seconds": 0.32570742948432496,
+    "producer_p95_seconds": 0.2170440009995218,
+    "gpu_update_seconds": 0.970857335509617,
+    "producer_batches_per_second": 3.070240066624351,
+    "gpu_batches_per_second": 1.0300174530535793,
+    "producer_headroom": 2.9807650904577865,
     "sustainable": true,
-    "buffer_seconds": 60.10053018116257,
+    "buffer_seconds": 60.19315480159625,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 215513.0633967323,
-    "upload_bytes_per_second": 5047331.910905213,
+    "generation_bytes_per_second": 210356.95741280887,
+    "upload_bytes_per_second": 6654698.852183321,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 46.24408651758648,
+    "estimated_full_shards_per_hour": 45.13770620144081,
     "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": 22.672299763999945,
-      "bytes": 7363447,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.0496788179998475,
       "bytes": 15965043,
@@ -562,6 +556,12 @@
       "bytes": 1132593,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 0.9999866519992793,
+      "bytes": 6918002,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -90,3 +90,4 @@
 {"step": 9000, "loss": 0.7084479369223118, "lr": 0.0003, "grad_norm": 0.15896695598011765, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1316240288, "tokens_per_second": 149722.04308527088, "data_wait_s": 1.7378000848111697e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9100, "loss": 0.6799753941595554, "lr": 0.0003, "grad_norm": 0.15308886010203326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1330706081, "tokens_per_second": 146366.16450904598, "data_wait_s": 1.2032998711219989e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 9200, "loss": 0.8535147793591022, "lr": 0.0003, "grad_norm": 0.33993378266042257, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1345371939, "tokens_per_second": 150043.87223557767, "data_wait_s": 1.4172999726724811e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 9300, "loss": 0.7113298326730728, "lr": 0.0003, "grad_norm": 0.1712786023034142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 1360127552, "tokens_per_second": 150064.52455482446, "data_wait_s": 1.2900998626719229e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:082102411f2d738d739d15cdeeb97d98f9d7465464716de6b9748e509837f6d0
-size 192535
+oid sha256:4c2bd70db2fa28486d728fed199341b2c7d07dc12f1f017debf2afd468d9280f
+size 194618