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": 74691.14644282627,
-      "required_pages_per_second_upper_estimate": 0.006474039270684214
+      "required_token_positions_per_second": 74494.6071233737,
+      "required_pages_per_second_upper_estimate": 0.006457003740598416
     },
     {
       "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": 9336.393305353284,
-      "required_pages_per_second_upper_estimate": 0.0037244268810249256
+      "required_token_positions_per_second": 9311.825890421713,
+      "required_pages_per_second_upper_estimate": 0.0037146265718931357
     },
     {
       "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": 18672.786610706567,
-      "required_pages_per_second_upper_estimate": 0.004319620552566718
+      "required_token_positions_per_second": 18623.651780843426,
+      "required_pages_per_second_upper_estimate": 0.004308254074421351
     },
     {
       "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": 28009.179916059853,
-      "required_pages_per_second_upper_estimate": 0.001705612104078137
+      "required_token_positions_per_second": 27935.47767126514,
+      "required_pages_per_second_upper_estimate": 0.0017011240240559328
     },
     {
       "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": 28009.179916059853,
-      "required_pages_per_second_upper_estimate": 0.002156224204310716
+      "required_token_positions_per_second": 27935.47767126514,
+      "required_pages_per_second_upper_estimate": 0.0021505504014855473
     },
     {
       "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": 28009.179916059853,
-      "required_pages_per_second_upper_estimate": 0.3477802738624465
+      "required_token_positions_per_second": 27935.47767126514,
+      "required_pages_per_second_upper_estimate": 0.34686513864764196
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.12368116671080998,
-    "producer_p95_seconds": 0.33008311100275023,
-    "gpu_update_seconds": 0.9871049449808147,
-    "producer_batches_per_second": 8.085305358884508,
-    "gpu_batches_per_second": 1.013063509695452,
-    "producer_headroom": 7.981044901434777,
+    "producer_batch_seconds": 0.09980742682822097,
+    "producer_p95_seconds": 0.3139325949996419,
+    "gpu_update_seconds": 0.9897092265738902,
+    "producer_batches_per_second": 10.019294473156839,
+    "gpu_batches_per_second": 1.0103977745683284,
+    "producer_headroom": 9.916188183844108,
     "sustainable": true,
-    "buffer_seconds": 60.2134016438297,
+    "buffer_seconds": 60.3722628210073,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 666676.4562270228,
+    "generation_bytes_per_second": 641663.6209061916,
     "upload_bytes_per_second": 4863237.997411955,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.52662403635031,
+    "estimated_full_shards_per_hour": 68.84303794092804,
     "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-v1/1790782192342215557-281f6532/train.jsonl

@@ -275,3 +275,4 @@
 {"step": 382500, "loss": 0.587795801460743, "lr": 0.0003, "grad_norm": 0.17074154889065904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55986664145, "tokens_per_second": 148913.49731703813, "data_wait_s": 1.3286000466905534e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382600, "loss": 0.6028595305979252, "lr": 0.0003, "grad_norm": 0.17578412955566114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56001388975, "tokens_per_second": 150112.0015920155, "data_wait_s": 1.0195999493589625e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382700, "loss": 0.5502717569470406, "lr": 0.0003, "grad_norm": 0.1383424712148887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56016046452, "tokens_per_second": 148644.20057002187, "data_wait_s": 7.5600000855047256e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 382800, "loss": 0.5965701099485159, "lr": 0.0003, "grad_norm": 0.15886674502314926, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56030838053, "tokens_per_second": 151526.8503122318, "data_wait_s": 1.5002999134594575e-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 sha256:157f09970cf15fab518c656fef661bfe14956fdef83b3371fb8a63de70610eb3
-size 590766
+oid sha256:8038ec3d84292dc4e5e463a2f9e4c5271321d137cd861ab711251ba0726592e0
+size 592884