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": 74851.18918937142,
-      "required_pages_per_second_upper_estimate": 0.0064879113703301145
+      "required_token_positions_per_second": 74892.44031452967,
+      "required_pages_per_second_upper_estimate": 0.006491486913308811
     },
     {
       "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": 9356.398648671428,
-      "required_pages_per_second_upper_estimate": 0.0037324073115810705
+      "required_token_positions_per_second": 9361.555039316208,
+      "required_pages_per_second_upper_estimate": 0.003734464272904184
     },
     {
       "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": 18712.797297342855,
-      "required_pages_per_second_upper_estimate": 0.004328876320755988
+      "required_token_positions_per_second": 18723.110078632417,
+      "required_pages_per_second_upper_estimate": 0.004331262001208577
     },
     {
       "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": 28069.195946014283,
-      "required_pages_per_second_upper_estimate": 0.0017092667654225878
+      "required_token_positions_per_second": 28084.665117948625,
+      "required_pages_per_second_upper_estimate": 0.0017102087568329166
     },
     {
       "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": 28069.195946014283,
-      "required_pages_per_second_upper_estimate": 0.0021608444044316118
+      "required_token_positions_per_second": 28084.665117948625,
+      "required_pages_per_second_upper_estimate": 0.002162035264108526
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28069.195946014283,
-      "required_pages_per_second_upper_estimate": 0.34852547209374923
+      "required_token_positions_per_second": 28084.665117948625,
+      "required_pages_per_second_upper_estimate": 0.3487175474371857
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.05930502062494725,
-    "producer_p95_seconds": 0.21414185399953567,
-    "gpu_update_seconds": 0.9849943708104653,
-    "producer_batches_per_second": 16.86197879137639,
-    "gpu_batches_per_second": 1.0152342283714657,
-    "producer_headroom": 16.6089541902312,
+    "producer_batch_seconds": 0.10284103876562511,
+    "producer_p95_seconds": 0.20098331999997754,
+    "gpu_update_seconds": 0.9844518310574565,
+    "producer_batches_per_second": 9.723744645160592,
+    "gpu_batches_per_second": 1.0157937325646926,
+    "producer_headroom": 9.572558220663481,
     "sustainable": true,
-    "buffer_seconds": 60.084656619438384,
+    "buffer_seconds": 60.05156169450485,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369186.11055891396,
-    "upload_bytes_per_second": 5937973.585928265,
+    "generation_bytes_per_second": 368746.4906034378,
+    "upload_bytes_per_second": 6012327.911845041,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.21874511313976,
+    "estimated_full_shards_per_hour": 79.12441290452338,
     "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": [
@@ -544,6 +544,12 @@
       "bytes": 5552102,
       "pages": 19,
       "shards": 2
+    },
+    {
+      "seconds": 1.2703921679994892,
+      "bytes": 12396222,
+      "pages": 47,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790476562033999719-ba4f58e1/train.jsonl

@@ -71,3 +71,4 @@
 {"step": 89000, "loss": 0.6425842028111219, "lr": 0.0003, "grad_norm": 0.15072500295912583, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13024704107, "tokens_per_second": 149202.7943761418, "data_wait_s": 1.308999981119996e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 89100, "loss": 0.5993454270064831, "lr": 0.0003, "grad_norm": 0.13591362712625177, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13039297606, "tokens_per_second": 145374.30608773872, "data_wait_s": 1.313800021307543e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 89200, "loss": 0.6013900879770517, "lr": 0.0003, "grad_norm": 0.18664570087347193, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13053837626, "tokens_per_second": 148020.184680706, "data_wait_s": 1.192800027638441e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 89300, "loss": 0.6066636741161346, "lr": 0.0003, "grad_norm": 0.12688301071404934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 13068424989, "tokens_per_second": 148636.4784598391, "data_wait_s": 1.2800000149582047e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790476562.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.417.0.1790476562033999719-ba4f58e1

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:30da922a0a42c2ebc4b011db2a7b0efcfa6f354980d6bc3a72c949861f6fe449
-size 155562
+oid sha256:b3d83f5e401f1554f99a406ae2a795f3433d52a42ace2e8b74caa9955d248ed6
+size 157680