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": 77220.68572185696,
-      "required_pages_per_second_upper_estimate": 0.005342306214914668
+      "required_token_positions_per_second": 76353.86260436993,
+      "required_pages_per_second_upper_estimate": 0.005282337380340178
     },
     {
       "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": 9652.58571523212,
-      "required_pages_per_second_upper_estimate": 0.00404133607005679
+      "required_token_positions_per_second": 9544.232825546242,
+      "required_pages_per_second_upper_estimate": 0.003995970977810946
     },
     {
       "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": 19305.17143046424,
-      "required_pages_per_second_upper_estimate": 0.004020758097358807
+      "required_token_positions_per_second": 19088.465651092483,
+      "required_pages_per_second_upper_estimate": 0.003975623998431384
     },
     {
       "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": 28957.757145696363,
-      "required_pages_per_second_upper_estimate": 0.001735973324949521
+      "required_token_positions_per_second": 28632.69847663872,
+      "required_pages_per_second_upper_estimate": 0.0017164865540753645
     },
     {
       "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": 28957.757145696363,
-      "required_pages_per_second_upper_estimate": 0.0022513382463652657
+      "required_token_positions_per_second": 28632.69847663872,
+      "required_pages_per_second_upper_estimate": 0.0022260663646280154
     },
     {
       "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": 28957.757145696363,
-      "required_pages_per_second_upper_estimate": 0.3714390162478209
+      "required_token_positions_per_second": 28632.69847663872,
+      "required_pages_per_second_upper_estimate": 0.3672695126619556
     }
   ],
   "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": 0.11025657638302278,
-    "producer_p95_seconds": 0.2322854100020777,
-    "gpu_update_seconds": 0.9547700763181856,
-    "producer_batches_per_second": 9.069753776192703,
-    "gpu_batches_per_second": 1.0473725819479298,
-    "producer_headroom": 8.65952950508266,
+    "producer_batch_seconds": 0.06634789647665684,
+    "producer_p95_seconds": 0.2307602719993156,
+    "gpu_update_seconds": 0.965609302335156,
+    "producer_batches_per_second": 15.07206788917309,
+    "gpu_batches_per_second": 1.035615540966389,
+    "producer_headroom": 14.553728959212535,
     "sustainable": true,
-    "buffer_seconds": 60.15051480804569,
+    "buffer_seconds": 60.83338604711483,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 347050.47687860526,
-    "upload_bytes_per_second": 4773889.065568077,
+    "generation_bytes_per_second": 307114.7035336407,
+    "upload_bytes_per_second": 4827404.973646758,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.46895341652507,
+    "estimated_full_shards_per_hour": 65.89966611391941,
     "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": 1.0071225869996852,
-      "bytes": 672738,
-      "pages": 21,
-      "shards": 1
-    },
     {
       "seconds": 1.3492830719987978,
       "bytes": 12380695,
@@ -562,6 +556,12 @@
       "bytes": 7522148,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.2877170510000724,
+      "bytes": 6767236,
+      "pages": 27,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -186,3 +186,4 @@
 {"step": 415700, "loss": 0.5602020025253296, "lr": 0.0003, "grad_norm": 0.150553616427438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60847417908, "tokens_per_second": 152569.99292592288, "data_wait_s": 1.350199818261899e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 415800, "loss": 0.6331706643104553, "lr": 0.0003, "grad_norm": 0.1568104825987768, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60862157194, "tokens_per_second": 153088.32221817953, "data_wait_s": 1.1731997801689431e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 415900, "loss": 0.5758548378944397, "lr": 0.0003, "grad_norm": 0.15755216020757526, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60876652439, "tokens_per_second": 117081.74765913683, "data_wait_s": 9.520001185592264e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 416000, "loss": 0.6045684814453125, "lr": 0.0003, "grad_norm": 0.15525646490416434, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60891258285, "tokens_per_second": 151928.47331500877, "data_wait_s": 1.3898999895900488e-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 sha256:338b1477f121b03f9ecd3ea6e5e901984d7b1af15256e4619ba9625a39a0b313
-size 401481
+oid sha256:1acb05c7b6b00cd2c38e250171479a6d2981ea08e3cd2d16a18e1618bf2ca89f
+size 403599