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": 76935.68577200595,
-      "required_pages_per_second_upper_estimate": 0.005322589257092983
+      "required_token_positions_per_second": 76459.10548059428,
+      "required_pages_per_second_upper_estimate": 0.005289618326714496
     },
     {
       "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": 9616.960721500744,
-      "required_pages_per_second_upper_estimate": 0.004026420629115927
+      "required_token_positions_per_second": 9557.388185074286,
+      "required_pages_per_second_upper_estimate": 0.004001478852130191
     },
     {
       "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": 19233.921443001487,
-      "required_pages_per_second_upper_estimate": 0.004005918603958842
+      "required_token_positions_per_second": 19114.77637014857,
+      "required_pages_per_second_upper_estimate": 0.003981103827350445
     },
     {
       "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": 28850.88216450223,
-      "required_pages_per_second_upper_estimate": 0.0017295663330155803
+      "required_token_positions_per_second": 28672.164555222855,
+      "required_pages_per_second_upper_estimate": 0.0017188524852252704
     },
     {
       "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": 28850.88216450223,
-      "required_pages_per_second_upper_estimate": 0.002243029186670783
+      "required_token_positions_per_second": 28672.164555222855,
+      "required_pages_per_second_upper_estimate": 0.002229134678644998
     },
     {
       "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": 28850.88216450223,
-      "required_pages_per_second_upper_estimate": 0.37006813874247674
+      "required_token_positions_per_second": 28672.164555222855,
+      "required_pages_per_second_upper_estimate": 0.3677757411426592
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.12971971186701126,
-    "producer_p95_seconds": 0.2836799949982378,
-    "gpu_update_seconds": 0.9583069190867848,
-    "producer_batches_per_second": 7.708928624704322,
-    "gpu_batches_per_second": 1.0435070227322856,
-    "producer_headroom": 7.387519639800324,
+    "producer_batch_seconds": 0.3487495907109235,
+    "producer_p95_seconds": 0.2760251719992084,
+    "gpu_update_seconds": 0.964280180059294,
+    "producer_batches_per_second": 2.867386877677784,
+    "gpu_batches_per_second": 1.037042988831845,
+    "producer_headroom": 2.7649643347067903,
     "sustainable": true,
-    "buffer_seconds": 60.37333590246744,
+    "buffer_seconds": 60.74965134373552,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 456947.74067861313,
+    "generation_bytes_per_second": 408405.0600899362,
     "upload_bytes_per_second": 5244926.270432152,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.05034795063777,
+    "estimated_full_shards_per_hour": 87.63421871207775,
     "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": [

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -287,3 +287,4 @@
 {"step": 633100, "loss": 0.5014548301696777, "lr": 0.0001, "grad_norm": 0.1663041007092924, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92668748003, "tokens_per_second": 150633.27313737542, "data_wait_s": 2.2201998945092782e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 633200, "loss": 0.5431302785873413, "lr": 0.0001, "grad_norm": 0.1820425139968099, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92683388218, "tokens_per_second": 151860.21411178264, "data_wait_s": 1.1710999388014898e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 633300, "loss": 0.5552104711532593, "lr": 0.0001, "grad_norm": 0.15762070556440053, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92698069634, "tokens_per_second": 152304.11420733147, "data_wait_s": 1.6681999113643542e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 633400, "loss": 0.548349916934967, "lr": 0.0001, "grad_norm": 0.1737751292020463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92712979743, "tokens_per_second": 154007.99723675137, "data_wait_s": 1.6017002053558826e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d4fc0c1b8b67a15bb1bf8eab986a24eec43079c8fcd82d90d81af9357ed884b1
-size 616965
+oid sha256:98af1af089c74c8fe01b5776c575b49e64c809c46946f8ad61f45e75711eec8f
+size 619083