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": 77255.45707424024,
-      "required_pages_per_second_upper_estimate": 0.005344711777753196
+      "required_token_positions_per_second": 77253.0819505995,
+      "required_pages_per_second_upper_estimate": 0.0053445474611886834
     },
     {
       "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": 9656.93213428003,
-      "required_pages_per_second_upper_estimate": 0.004043155824948599
+      "required_token_positions_per_second": 9656.635243824938,
+      "required_pages_per_second_upper_estimate": 0.004043031523114829
     },
     {
       "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": 19313.86426856006,
-      "required_pages_per_second_upper_estimate": 0.00402256858628861
+      "required_token_positions_per_second": 19313.270487649876,
+      "required_pages_per_second_upper_estimate": 0.004022444917384074
     },
     {
       "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": 28970.79640284009,
-      "required_pages_per_second_upper_estimate": 0.0017367550085054947
+      "required_token_positions_per_second": 28969.90573147481,
+      "required_pages_per_second_upper_estimate": 0.0017367016141171226
     },
     {
       "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": 28970.79640284009,
-      "required_pages_per_second_upper_estimate": 0.00225235199125456
+      "required_token_positions_per_second": 28969.90573147481,
+      "required_pages_per_second_upper_estimate": 0.0022522827454735587
     },
     {
       "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": 28970.79640284009,
-      "required_pages_per_second_upper_estimate": 0.37160626983799705
+      "required_token_positions_per_second": 28969.90573147481,
+      "required_pages_per_second_upper_estimate": 0.37159484526205167
     }
   ],
   "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": 0.08644144331250914,
-    "producer_p95_seconds": 0.1866955430004964,
-    "gpu_update_seconds": 0.9543403507295225,
-    "producer_batches_per_second": 11.568525023173551,
-    "gpu_batches_per_second": 1.0478441985980935,
-    "producer_headroom": 11.040310228038704,
+    "producer_batch_seconds": 0.32447631323434933,
+    "producer_p95_seconds": 0.18748494000010396,
+    "gpu_update_seconds": 0.9543696916473358,
+    "producer_batches_per_second": 3.081889060042917,
+    "gpu_batches_per_second": 1.0478119839219766,
+    "producer_headroom": 2.941261511924456,
     "sustainable": true,
-    "buffer_seconds": 60.123442095959916,
+    "buffer_seconds": 60.12529057378215,
     "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": 457598.1451545913,
-    "upload_bytes_per_second": 5454608.15845552,
+    "generation_bytes_per_second": 410726.2539121909,
+    "upload_bytes_per_second": 5451612.306446722,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.18990961054138,
+    "estimated_full_shards_per_hour": 88.13229287170692,
     "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": [
@@ -538,6 +538,12 @@
       "bytes": 7663876,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 0.9853112940008941,
+      "bytes": 5162186,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -70,3 +70,4 @@
 {"step": 611400, "loss": 0.5387265086174011, "lr": 0.0001, "grad_norm": 0.14354600996511957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89490991957, "tokens_per_second": 152278.71618694995, "data_wait_s": 1.7242000467376783e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 611500, "loss": 0.5263700485229492, "lr": 0.0001, "grad_norm": 0.1665427354378426, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89505662878, "tokens_per_second": 152567.99240547285, "data_wait_s": 7.689000085520092e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 611600, "loss": 0.5925382971763611, "lr": 0.0001, "grad_norm": 0.16367374215516606, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89520455201, "tokens_per_second": 154400.72257098564, "data_wait_s": 1.4102000022830907e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 611700, "loss": 0.5162065029144287, "lr": 0.0001, "grad_norm": 0.17406104491443863, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89535262022, "tokens_per_second": 153441.24196737166, "data_wait_s": 1.3316000149643514e-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 sha256:1732fba8e704a15a3bfd566083da0f5dd3605dfd627c2380aba7e38f36122dda
-size 154227
+oid sha256:f40ac973924338f699a5fb1469d922f148734fea189839412d4594a2501bb34d
+size 156345