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": 77494.35828163257,
-      "required_pages_per_second_upper_estimate": 0.005361239517607769
+      "required_token_positions_per_second": 77570.33896284038,
+      "required_pages_per_second_upper_estimate": 0.005366496037433196
     },
     {
       "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": 9686.79478520407,
-      "required_pages_per_second_upper_estimate": 0.004055658693287431
+      "required_token_positions_per_second": 9696.292370355048,
+      "required_pages_per_second_upper_estimate": 0.004059635133858014
     },
     {
       "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": 19373.58957040814,
-      "required_pages_per_second_upper_estimate": 0.004035007791601437
+      "required_token_positions_per_second": 19392.584740710096,
+      "required_pages_per_second_upper_estimate": 0.004038963984639007
     },
     {
       "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": 29060.384355612212,
-      "required_pages_per_second_upper_estimate": 0.0017421256694813002
+      "required_token_positions_per_second": 29088.877111065143,
+      "required_pages_per_second_upper_estimate": 0.0017438337666647848
     },
     {
       "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": 29060.384355612212,
-      "required_pages_per_second_upper_estimate": 0.002259317060526833
+      "required_token_positions_per_second": 29088.877111065143,
+      "required_pages_per_second_upper_estimate": 0.002261532246936912
     },
     {
       "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": 29060.384355612212,
-      "required_pages_per_second_upper_estimate": 0.3727554079041086
+      "required_token_positions_per_second": 29088.877111065143,
+      "required_pages_per_second_upper_estimate": 0.37312088237792157
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.0836982888202158,
-    "producer_p95_seconds": 0.18629508199956035,
-    "gpu_update_seconds": 0.9513982905962685,
-    "producer_batches_per_second": 11.947675562973615,
-    "gpu_batches_per_second": 1.0510845036028722,
-    "producer_headroom": 11.366998107211908,
+    "producer_batch_seconds": 0.4685807674609208,
+    "producer_p95_seconds": 0.18535798700031592,
+    "gpu_update_seconds": 0.9504663894187567,
+    "producer_batches_per_second": 2.1341038075861682,
+    "gpu_batches_per_second": 1.0521150575472056,
+    "producer_headroom": 2.0283939406412466,
     "sustainable": true,
-    "buffer_seconds": 60.88949059816118,
+    "buffer_seconds": 60.82984892280043,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340207.645984184,
-    "upload_bytes_per_second": 4113683.5644215327,
+    "generation_bytes_per_second": 320404.672807767,
+    "upload_bytes_per_second": 4113095.128399541,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.0006411995329,
+    "estimated_full_shards_per_hour": 68.75138414549596,
     "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": [
@@ -412,6 +412,12 @@
       "bytes": 11665056,
       "pages": 37,
       "shards": 3
+    },
+    {
+      "seconds": 1.3450324119985453,
+      "bytes": 5494810,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -88,3 +88,4 @@
 {"step": 933800, "loss": 0.5353503227233887, "lr": 3e-05, "grad_norm": 0.3380934524447216, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136681821891, "tokens_per_second": 152289.93964229143, "data_wait_s": 7.972001185407862e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 933900, "loss": 0.5196212530136108, "lr": 3e-05, "grad_norm": 0.33102221519367003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136696449396, "tokens_per_second": 153363.11647315358, "data_wait_s": 1.1234000339754857e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 934000, "loss": 0.522619366645813, "lr": 3e-05, "grad_norm": 0.33103265439524876, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136711042179, "tokens_per_second": 142694.9714276392, "data_wait_s": 1.3750999642070383e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 934100, "loss": 0.557944655418396, "lr": 3e-05, "grad_norm": 0.3649112209828441, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 136725771571, "tokens_per_second": 153690.23401889257, "data_wait_s": 1.0957000995404087e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cfa8762d740b84468820806ffeae204daee0f6feb5bb205f93efa63c4d06308d
-size 191568
+oid sha256:da7691b18d6c55c08b178b8d6e775a08d36fd33b93da52442861518ecfa8f197
+size 193686