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": 74616.67922449438,
-      "required_pages_per_second_upper_estimate": 0.006467584640934645
+      "required_token_positions_per_second": 74783.973714564,
+      "required_pages_per_second_upper_estimate": 0.00648208530332987
     },
     {
       "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": 9327.084903061797,
-      "required_pages_per_second_upper_estimate": 0.0037207136201778354
+      "required_token_positions_per_second": 9347.9967143205,
+      "required_pages_per_second_upper_estimate": 0.003729055654348372
     },
     {
       "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": 18654.169806123595,
-      "required_pages_per_second_upper_estimate": 0.004315313882471018
+      "required_token_positions_per_second": 18695.993428641,
+      "required_pages_per_second_upper_estimate": 0.004324989041469803
     },
     {
       "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": 27981.25470918539,
-      "required_pages_per_second_upper_estimate": 0.0017039116054917195
+      "required_token_positions_per_second": 28043.9901429615,
+      "required_pages_per_second_upper_estimate": 0.0017077318642613015
     },
     {
       "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": 27981.25470918539,
-      "required_pages_per_second_upper_estimate": 0.0021540744445835993
+      "required_token_positions_per_second": 28043.9901429615,
+      "required_pages_per_second_upper_estimate": 0.00215890399193848
     },
     {
       "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": 27981.25470918539,
-      "required_pages_per_second_upper_estimate": 0.34743353625272094
+      "required_token_positions_per_second": 28043.9901429615,
+      "required_pages_per_second_upper_estimate": 0.34821250037822987
     }
   ],
   "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": 0.4054135304687634,
-    "producer_p95_seconds": 0.24095169599968358,
-    "gpu_update_seconds": 0.9880900727058536,
-    "producer_batches_per_second": 2.466617231160835,
-    "gpu_batches_per_second": 1.012053483405143,
-    "producer_headroom": 2.4372399992752207,
+    "producer_batch_seconds": 0.059340701453081124,
+    "producer_p95_seconds": 0.20147364500007825,
+    "gpu_update_seconds": 0.9858796789992139,
+    "producer_batches_per_second": 16.851839892567995,
+    "gpu_batches_per_second": 1.0143225601476236,
+    "producer_headroom": 16.613886503831083,
     "sustainable": true,
-    "buffer_seconds": 60.27349443505707,
+    "buffer_seconds": 60.13866041895205,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 316621.66056762595,
-    "upload_bytes_per_second": 5249086.6351809865,
+    "generation_bytes_per_second": 315655.06485452,
+    "upload_bytes_per_second": 5234108.162709917,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.93963778277954,
+    "estimated_full_shards_per_hour": 67.7322288439436,
     "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": [
@@ -460,6 +460,12 @@
       "bytes": 5906480,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.2417195279995212,
+      "bytes": 5623378,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -85,3 +85,4 @@
 {"step": 203500, "loss": 0.5775925777852535, "lr": 0.0003, "grad_norm": 0.1425142528755673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29784752155, "tokens_per_second": 147981.58452242424, "data_wait_s": 1.9011000404134393e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203600, "loss": 0.5829173810780048, "lr": 0.0003, "grad_norm": 0.15844119669134157, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29799460784, "tokens_per_second": 148556.21699054367, "data_wait_s": 1.7874001059681177e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 203700, "loss": 0.6329334229230881, "lr": 0.0003, "grad_norm": 0.1702569485568475, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29814096252, "tokens_per_second": 147765.11220061584, "data_wait_s": 1.8253998860018328e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 203800, "loss": 0.5466072801500559, "lr": 0.0003, "grad_norm": 0.14604799335517452, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 29828659819, "tokens_per_second": 146933.57037424322, "data_wait_s": 2.0323001081123948e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6017420b097520ca60712adbcaf4614d96a979759a58ed4ca084e7e5fb9cac7e
-size 185214
+oid sha256:ef1d2413b2f396b9fbd69097ba72e485950ef642a8052c5990dcd7a278044368
+size 187332