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": 76876.7275407287,
-      "required_pages_per_second_upper_estimate": 0.005318510389851278
+      "required_token_positions_per_second": 77088.77378732488,
+      "required_pages_per_second_upper_estimate": 0.0053331802412058815
     },
     {
       "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": 9609.590942591087,
-      "required_pages_per_second_upper_estimate": 0.0040233350565849375
+      "required_token_positions_per_second": 9636.09672341561,
+      "required_pages_per_second_upper_estimate": 0.004034432473512521
     },
     {
       "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": 19219.181885182174,
-      "required_pages_per_second_upper_estimate": 0.004002848742773357
+      "required_token_positions_per_second": 19272.19344683122,
+      "required_pages_per_second_upper_estimate": 0.004013889653055961
     },
     {
       "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": 28828.77282777326,
-      "required_pages_per_second_upper_estimate": 0.001728240912037681
+      "required_token_positions_per_second": 28908.290170246833,
+      "required_pages_per_second_upper_estimate": 0.00173300785530302
     },
     {
       "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": 28828.77282777326,
-      "required_pages_per_second_upper_estimate": 0.0022413102829888006
+      "required_token_positions_per_second": 28908.290170246833,
+      "required_pages_per_second_upper_estimate": 0.0022474924066062953
     },
     {
       "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": 28828.77282777326,
-      "required_pages_per_second_upper_estimate": 0.36978454391007376
+      "required_token_positions_per_second": 28908.290170246833,
+      "required_pages_per_second_upper_estimate": 0.3708045070002544
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.08344079933613102,
-    "producer_p95_seconds": 0.3010225870020804,
-    "gpu_update_seconds": 0.9590418629739342,
-    "producer_batches_per_second": 11.9845448264658,
-    "gpu_batches_per_second": 1.0427073505415676,
-    "producer_headroom": 11.493680197268386,
+    "producer_batch_seconds": 0.1032613260388473,
+    "producer_p95_seconds": 0.31255616300040856,
+    "gpu_update_seconds": 0.9564038494554771,
+    "producer_batches_per_second": 9.684167716612473,
+    "gpu_batches_per_second": 1.0455834118289509,
+    "producer_headroom": 9.261975282940627,
     "sustainable": true,
-    "buffer_seconds": 60.41963736735786,
+    "buffer_seconds": 60.253442515695056,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321807.61304015084,
+    "generation_bytes_per_second": 321557.2081063102,
     "upload_bytes_per_second": 5876297.485687808,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.0524224605884,
+    "estimated_full_shards_per_hour": 68.99869139091473,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -303,3 +303,4 @@
 {"step": 427400, "loss": 0.5921719670295715, "lr": 0.0003, "grad_norm": 0.16026334633541897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62559354053, "tokens_per_second": 151263.85183425818, "data_wait_s": 1.1219999578315765e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427500, "loss": 0.5645163059234619, "lr": 0.0003, "grad_norm": 0.18016948607630323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62573916972, "tokens_per_second": 150934.04924435352, "data_wait_s": 1.7066999134840444e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 427600, "loss": 0.5873361229896545, "lr": 0.0003, "grad_norm": 0.17662522692257182, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62588560943, "tokens_per_second": 152404.64430579793, "data_wait_s": 1.518800127087161e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 427700, "loss": 0.6278319954872131, "lr": 0.0003, "grad_norm": 0.14338572818608436, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 62603273646, "tokens_per_second": 152269.30883924387, "data_wait_s": 8.983999578049406e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ad920f99d4351971c0db37ed97e603242b9d86a1ffbc70ff018f7802cc55a2b2
-size 650853
+oid sha256:9e217366454089c322e564f97ec133c1437048a8263f5e72f84b87f51e144e6b
+size 652971