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": 76753.2186925252,
-      "required_pages_per_second_upper_estimate": 0.00530996576115255
+      "required_token_positions_per_second": 77332.06377960955,
+      "required_pages_per_second_upper_estimate": 0.005350011607382691
     },
     {
       "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": 9594.15233656565,
-      "required_pages_per_second_upper_estimate": 0.004016871234636842
+      "required_token_positions_per_second": 9666.507972451194,
+      "required_pages_per_second_upper_estimate": 0.004047165028424626
     },
     {
       "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": 19188.3046731313,
-      "required_pages_per_second_upper_estimate": 0.003996417833790001
+      "required_token_positions_per_second": 19333.01594490239,
+      "required_pages_per_second_upper_estimate": 0.0040265573754070475
     },
     {
       "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": 28782.457009696947,
-      "required_pages_per_second_upper_estimate": 0.0017254643494641654
+      "required_token_positions_per_second": 28999.52391735358,
+      "required_pages_per_second_upper_estimate": 0.0017384771791362037
     },
     {
       "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": 28782.457009696947,
-      "required_pages_per_second_upper_estimate": 0.002237709431855121
+      "required_token_positions_per_second": 28999.52391735358,
+      "required_pages_per_second_upper_estimate": 0.0022545854291490014
     },
     {
       "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": 28782.457009696947,
-      "required_pages_per_second_upper_estimate": 0.36919045432584174
+      "required_token_positions_per_second": 28999.52391735358,
+      "required_pages_per_second_upper_estimate": 0.37197475554897425
     }
   ],
   "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": 0.05870067844530524,
-    "producer_p95_seconds": 0.19776052800079924,
-    "gpu_update_seconds": 0.9605851227602027,
-    "producer_batches_per_second": 17.035578233252565,
-    "gpu_batches_per_second": 1.0410321545752659,
-    "producer_headroom": 16.36412300847995,
+    "producer_batch_seconds": 0.11765826124209866,
+    "producer_p95_seconds": 0.21435111599930678,
+    "gpu_update_seconds": 0.9533949618895358,
+    "producer_batches_per_second": 8.49919070232015,
+    "gpu_batches_per_second": 1.0488832435385411,
+    "producer_headroom": 8.103085595730414,
     "sustainable": true,
-    "buffer_seconds": 60.51686273389277,
+    "buffer_seconds": 60.06388259904075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328753.0591787303,
+    "generation_bytes_per_second": 361584.7283551106,
     "upload_bytes_per_second": 3566608.9111582045,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.5427535202163,
+    "estimated_full_shards_per_hour": 77.58766544332494,
     "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-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -149,3 +149,4 @@
 {"step": 939900, "loss": 0.49588456749916077, "lr": 3e-05, "grad_norm": 0.3256690265589337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137574954840, "tokens_per_second": 153068.80787155707, "data_wait_s": 1.1854999684146605e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 940000, "loss": 0.5351184010505676, "lr": 3e-05, "grad_norm": 0.3404110635583141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137589403422, "tokens_per_second": 150593.29881988783, "data_wait_s": 1.485900065745227e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 940100, "loss": 0.5706033110618591, "lr": 3e-05, "grad_norm": 0.3362725201760242, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137603952938, "tokens_per_second": 143005.9843691961, "data_wait_s": 1.5850999261601828e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 940200, "loss": 0.4717559218406677, "lr": 3e-05, "grad_norm": 0.3242096837015167, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 137618600943, "tokens_per_second": 153029.22881389936, "data_wait_s": 1.1675001587718725e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1777df30228c2b1f9b4ecae1e74bf595f1f6e941110117cef57c573eb53dd9f3
-size 322332
+oid sha256:ffe9e80d35406ecdb8513e36a56f177e0c53569628d5dfdc61c6e644f80b4830
+size 324450