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": 75947.07305002237,
-      "required_pages_per_second_upper_estimate": 0.006582899805879087
+      "required_token_positions_per_second": 75999.01569776914,
+      "required_pages_per_second_upper_estimate": 0.0065874020629383395
     },
     {
       "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": 9493.384131252797,
-      "required_pages_per_second_upper_estimate": 0.0037870528686982595
+      "required_token_positions_per_second": 9499.876962221142,
+      "required_pages_per_second_upper_estimate": 0.003789642956048006
     },
     {
       "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": 18986.768262505593,
-      "required_pages_per_second_upper_estimate": 0.004392254681822083
+      "required_token_positions_per_second": 18999.753924442284,
+      "required_pages_per_second_upper_estimate": 0.0043952586861712925
     },
     {
       "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": 28480.152393758388,
-      "required_pages_per_second_upper_estimate": 0.0017342918569683563
+      "required_token_positions_per_second": 28499.630886663428,
+      "required_pages_per_second_upper_estimate": 0.0017354779949905193
     },
     {
       "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": 28480.152393758388,
-      "required_pages_per_second_upper_estimate": 0.00219248096937921
+      "required_token_positions_per_second": 28499.630886663428,
+      "required_pages_per_second_upper_estimate": 0.0021939804776829595
     },
     {
       "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": 28480.152393758388,
-      "required_pages_per_second_upper_estimate": 0.3536281757919762
+      "required_token_positions_per_second": 28499.630886663428,
+      "required_pages_per_second_upper_estimate": 0.35387003348353463
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.3532654386719969,
-    "producer_p95_seconds": 0.33748959799777367,
-    "gpu_update_seconds": 0.970781322295847,
-    "producer_batches_per_second": 2.830732617827608,
-    "gpu_batches_per_second": 1.0300981045196176,
-    "producer_headroom": 2.74802235380067,
+    "producer_batch_seconds": 0.12446394235178104,
+    "producer_p95_seconds": 0.3313772380024602,
+    "gpu_update_seconds": 0.9701178274887078,
+    "producer_batches_per_second": 8.034455450347465,
+    "gpu_batches_per_second": 1.0308026217687871,
+    "producer_headroom": 7.794368466545891,
     "sustainable": true,
-    "buffer_seconds": 60.18844198234252,
+    "buffer_seconds": 60.147305304299884,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415587.0658440933,
-    "upload_bytes_per_second": 3531657.2967227153,
+    "generation_bytes_per_second": 398909.96793208004,
+    "upload_bytes_per_second": 3555036.8511899454,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.1753099583826,
+    "estimated_full_shards_per_hour": 85.59679296943474,
     "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": 0.9224505880010838,
-      "bytes": 1117951,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0108905939996475,
       "bytes": 7845615,
@@ -562,6 +556,12 @@
       "bytes": 10633493,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.5226552149979398,
+      "bytes": 6060968,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -260,3 +260,4 @@
 {"step": 66000, "loss": 0.6565835401415825, "lr": 0.0003, "grad_norm": 0.1419560230238423, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9659487014, "tokens_per_second": 149725.03250765483, "data_wait_s": 1.1422998795751482e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 66100, "loss": 0.662840723991394, "lr": 0.0003, "grad_norm": 0.13320195560030293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9674131831, "tokens_per_second": 149037.0426219479, "data_wait_s": 1.156200232799165e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 66200, "loss": 0.6097782868891954, "lr": 0.0003, "grad_norm": 0.12760052589126836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9688766446, "tokens_per_second": 149353.62242378248, "data_wait_s": 1.0007002856582403e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 66300, "loss": 0.6145434500649571, "lr": 0.0003, "grad_norm": 0.14590196705388214, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9703281602, "tokens_per_second": 147787.86374603992, "data_wait_s": 1.2303997209528461e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1b1cc432eb3e045978b0fe389c6f4757bc7de4b0e9ed45073900b91b490b175
-size 558996
+oid sha256:11e0a4ffcb6a3891ab1dedf6b8a3ba9141d27f94530018481d649eba4820d690
+size 561114