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": 75923.4675427561,
-      "required_pages_per_second_upper_estimate": 0.006580853740336858
+      "required_token_positions_per_second": 75508.24926500011,
+      "required_pages_per_second_upper_estimate": 0.0065448636723820725
     },
     {
       "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": 9490.433442844513,
-      "required_pages_per_second_upper_estimate": 0.003785875794975472
+      "required_token_positions_per_second": 9438.531158125013,
+      "required_pages_per_second_upper_estimate": 0.0037651711975925537
     },
     {
       "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": 18980.866885689025,
-      "required_pages_per_second_upper_estimate": 0.004390889502156255
+      "required_token_positions_per_second": 18877.062316250027,
+      "required_pages_per_second_upper_estimate": 0.004366876141914569
     },
     {
       "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": 28471.300328533536,
-      "required_pages_per_second_upper_estimate": 0.0017337528126393597
+      "required_token_positions_per_second": 28315.59347437504,
+      "required_pages_per_second_upper_estimate": 0.0017242710821518356
     },
     {
       "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": 28471.300328533536,
-      "required_pages_per_second_upper_estimate": 0.002191799512894116
+      "required_token_positions_per_second": 28315.59347437504,
+      "required_pages_per_second_upper_estimate": 0.002179812768236836
     },
     {
       "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": 28471.300328533536,
-      "required_pages_per_second_upper_estimate": 0.3535182627678401
+      "required_token_positions_per_second": 28315.59347437504,
+      "required_pages_per_second_upper_estimate": 0.35158490475650994
     }
   ],
   "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": 0.0926331363671693,
-    "producer_p95_seconds": 0.3361694359991816,
-    "gpu_update_seconds": 0.9710831497320677,
-    "producer_batches_per_second": 10.795273043938696,
-    "gpu_batches_per_second": 1.029777934336427,
-    "producer_headroom": 10.483107749725674,
+    "producer_batch_seconds": 0.3518057271483599,
+    "producer_p95_seconds": 0.384082982003747,
+    "gpu_update_seconds": 0.9764231155889705,
+    "producer_batches_per_second": 2.842477887172912,
+    "gpu_batches_per_second": 1.0241461760118287,
+    "producer_headroom": 2.775461114586129,
     "sustainable": true,
-    "buffer_seconds": 60.2071552833882,
+    "buffer_seconds": 60.53823316651617,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 472094.94318714534,
-    "upload_bytes_per_second": 5256088.390285686,
+    "generation_bytes_per_second": 457924.55451891344,
+    "upload_bytes_per_second": 5328764.108006176,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 101.30058500014086,
+    "estimated_full_shards_per_hour": 98.2599494617038,
     "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": 0.9240367780002998,
-      "bytes": 5199805,
-      "pages": 10,
-      "shards": 2
-    },
     {
       "seconds": 1.4070672229972843,
       "bytes": 1010316,
@@ -562,6 +556,12 @@
       "bytes": 6495200,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1027901979978196,
+      "bytes": 11991791,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -327,3 +327,4 @@
 {"step": 147700, "loss": 0.6182952933013439, "lr": 0.0003, "grad_norm": 0.1372575641211036, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21615276738, "tokens_per_second": 148059.53962410265, "data_wait_s": 9.177994797937572e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 147800, "loss": 0.6207619551569223, "lr": 0.0003, "grad_norm": 0.21111374772444552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21629865402, "tokens_per_second": 149524.73886630964, "data_wait_s": 1.1203999747522175e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 147900, "loss": 0.617692619562149, "lr": 0.0003, "grad_norm": 0.13953698809683324, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21644514832, "tokens_per_second": 149675.14264660885, "data_wait_s": 1.4195000403560698e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 148000, "loss": 0.6084747966378927, "lr": 0.0003, "grad_norm": 0.15072616288645516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21659252572, "tokens_per_second": 151366.57806812413, "data_wait_s": 1.7004000255838037e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c0c2dc1e77d4d1df9a50439709374c2045406e5ba34d21c3f8af5c1a0837617
-size 701685
+oid sha256:d78b999cab081dc42d0a1f530622de9c8417f3422fbd5cf0a6e44dad6dd03547
+size 703803