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": 76712.45088976779,
-      "required_pages_per_second_upper_estimate": 0.005307145349963457
+      "required_token_positions_per_second": 76301.13600814914,
+      "required_pages_per_second_upper_estimate": 0.005278689632071066
     },
     {
       "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": 9589.056361220973,
-      "required_pages_per_second_upper_estimate": 0.004014737656176092
+      "required_token_positions_per_second": 9537.642001018643,
+      "required_pages_per_second_upper_estimate": 0.003993211537213307
     },
     {
       "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": 19178.112722441947,
-      "required_pages_per_second_upper_estimate": 0.003994295119241223
+      "required_token_positions_per_second": 19075.284002037286,
+      "required_pages_per_second_upper_estimate": 0.003972878608556649
     },
     {
       "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": 28767.16908366292,
-      "required_pages_per_second_upper_estimate": 0.001724547861641737
+      "required_token_positions_per_second": 28612.926003055927,
+      "required_pages_per_second_upper_estimate": 0.001715301224474895
     },
     {
       "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": 28767.16908366292,
-      "required_pages_per_second_upper_estimate": 0.0022365208628504552
+      "required_token_positions_per_second": 28612.926003055927,
+      "required_pages_per_second_upper_estimate": 0.002224529141776873
     },
     {
       "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": 28767.16908366292,
-      "required_pages_per_second_upper_estimate": 0.36899435722557333
+      "required_token_positions_per_second": 28612.926003055927,
+      "required_pages_per_second_upper_estimate": 0.3670158926008636
     }
   ],
   "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": 0.2529109868827675,
-    "producer_p95_seconds": 0.284690178999881,
-    "gpu_update_seconds": 0.9610956128353101,
-    "producer_batches_per_second": 3.95396029379907,
-    "gpu_batches_per_second": 1.0404792058616508,
-    "producer_headroom": 3.8001338916953,
+    "producer_batch_seconds": 0.0943237705468789,
+    "producer_p95_seconds": 0.27228982700034976,
+    "gpu_update_seconds": 0.9662765701434075,
+    "producer_batches_per_second": 10.60178144069209,
+    "gpu_batches_per_second": 1.0349003907355299,
+    "producer_headroom": 10.244253007921987,
     "sustainable": true,
-    "buffer_seconds": 60.54902360862454,
+    "buffer_seconds": 60.875423919034674,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 460729.26989528205,
+    "generation_bytes_per_second": 442705.0864262266,
     "upload_bytes_per_second": 5180278.733353433,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 98.8617760910401,
+    "estimated_full_shards_per_hour": 94.99420590009784,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -248,3 +248,4 @@
 {"step": 629200, "loss": 0.5503384470939636, "lr": 0.0001, "grad_norm": 0.17518068853209828, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92098434220, "tokens_per_second": 152921.2282267415, "data_wait_s": 1.6170000890269876e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 629300, "loss": 0.5630358457565308, "lr": 0.0001, "grad_norm": 0.15632178971311791, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92113039369, "tokens_per_second": 150502.20007090637, "data_wait_s": 1.5194000297924504e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 629400, "loss": 0.5516194701194763, "lr": 0.0001, "grad_norm": 0.1670101357463071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92127550988, "tokens_per_second": 150225.64819297375, "data_wait_s": 1.5941001038299873e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 629500, "loss": 0.5443993806838989, "lr": 0.0001, "grad_norm": 0.18413428447510263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92142069085, "tokens_per_second": 150035.13894966664, "data_wait_s": 1.759999940986745e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:97aa0e0e6db5084128c536eccacbffc14609004e2d13ae413388ab4e71de006e
-size 533580
+oid sha256:544c29bfa70614ebe691d566ddebfff29fa4a157d47a25c55a303054074e92fe
+size 535698