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": 75957.50502562102,
-      "required_pages_per_second_upper_estimate": 0.006583804023084378
+      "required_token_positions_per_second": 75100.37489180775,
+      "required_pages_per_second_upper_estimate": 0.006509510155461907
     },
     {
       "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": 9494.688128202628,
-      "required_pages_per_second_upper_estimate": 0.003787573052578039
+      "required_token_positions_per_second": 9387.546861475968,
+      "required_pages_per_second_upper_estimate": 0.0037448327993760844
     },
     {
       "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": 18989.376256405256,
-      "required_pages_per_second_upper_estimate": 0.004392857995311646
+      "required_token_positions_per_second": 18775.093722951937,
+      "required_pages_per_second_upper_estimate": 0.004343287502441005
     },
     {
       "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": 28484.064384607886,
-      "required_pages_per_second_upper_estimate": 0.001734530076686461
+      "required_token_positions_per_second": 28162.640584427907,
+      "required_pages_per_second_upper_estimate": 0.0017149570536358495
     },
     {
       "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": 28484.064384607886,
-      "required_pages_per_second_upper_estimate": 0.002192782125263888
+      "required_token_positions_per_second": 28162.640584427907,
+      "required_pages_per_second_upper_estimate": 0.0021680380313680073
     },
     {
       "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": 28484.064384607886,
-      "required_pages_per_second_upper_estimate": 0.3536767496257358
+      "required_token_positions_per_second": 28162.640584427907,
+      "required_pages_per_second_upper_estimate": 0.3496857417637596
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.26065642229644936,
-    "producer_p95_seconds": 0.36743696100165835,
-    "gpu_update_seconds": 0.9706479955487085,
-    "producer_batches_per_second": 3.8364679112440268,
-    "gpu_batches_per_second": 1.0302395972442087,
-    "producer_headroom": 3.723859888035955,
+    "producer_batch_seconds": 0.11529335192977896,
+    "producer_p95_seconds": 0.37913584600028116,
+    "gpu_update_seconds": 0.9817261246194198,
+    "producer_batches_per_second": 8.67352699233746,
+    "gpu_batches_per_second": 1.0186140257677918,
+    "producer_headroom": 8.515028040969385,
     "sustainable": true,
-    "buffer_seconds": 60.180175724019925,
+    "buffer_seconds": 60.867019726404024,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 401615.59699398594,
-    "upload_bytes_per_second": 5880139.674039189,
+    "generation_bytes_per_second": 401358.6861881517,
+    "upload_bytes_per_second": 5776158.800846744,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.17735798229869,
+    "estimated_full_shards_per_hour": 86.12223090394414,
     "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": 0.9420911959969089,
-      "bytes": 7110283,
-      "pages": 25,
-      "shards": 3
-    },
     {
       "seconds": 1.256377692996466,
       "bytes": 6950847,
@@ -562,6 +556,12 @@
       "bytes": 16969598,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.1678855200007092,
+      "bytes": 919043,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 396300, "loss": 0.5455999113619328, "lr": 0.0003, "grad_norm": 0.15656467796505655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58006849836, "tokens_per_second": 148805.33496927447, "data_wait_s": 1.1357995390426368e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 396400, "loss": 0.521472217515111, "lr": 0.0003, "grad_norm": 0.17155483494884516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58021582245, "tokens_per_second": 150481.0228060532, "data_wait_s": 1.0001000191550702e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 396500, "loss": 0.5753758829087019, "lr": 0.0003, "grad_norm": 0.1413881660238825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58036253337, "tokens_per_second": 149687.7418879604, "data_wait_s": 1.720000000204891e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 396600, "loss": 0.6008331198245287, "lr": 0.0003, "grad_norm": 0.1821427877255445, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 58050988572, "tokens_per_second": 150142.8073788414, "data_wait_s": 1.4903998817317188e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:acae88d361cdab18cfb31213c893696769ef29bfbdd45accdcf62f11394e5e01
-size 885399
+oid sha256:f7fde0c6e0b4ee4b65f09a3738bb8d15d4b17332fee24cc86bb9304127c2ac4b
+size 887517