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": 77182.32747721067,
-      "required_pages_per_second_upper_estimate": 0.005339652502546647
+      "required_token_positions_per_second": 77345.04463777818,
+      "required_pages_per_second_upper_estimate": 0.0053509096532705375
     },
     {
       "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": 9647.790934651333,
-      "required_pages_per_second_upper_estimate": 0.0040393285955540184
+      "required_token_positions_per_second": 9668.130579722272,
+      "required_pages_per_second_upper_estimate": 0.004047844380205133
     },
     {
       "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": 19295.581869302667,
-      "required_pages_per_second_upper_estimate": 0.004018760844662585
+      "required_token_positions_per_second": 19336.261159444544,
+      "required_pages_per_second_upper_estimate": 0.004027233268014116
     },
     {
       "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": 28943.372803954004,
-      "required_pages_per_second_upper_estimate": 0.001735111005625685
+      "required_token_positions_per_second": 29004.39173916682,
+      "required_pages_per_second_upper_estimate": 0.0017387689976211737
     },
     {
       "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": 28943.372803954004,
-      "required_pages_per_second_upper_estimate": 0.00225021992706483
+      "required_token_positions_per_second": 29004.39173916682,
+      "required_pages_per_second_upper_estimate": 0.0022549638808836954
     },
     {
       "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": 28943.372803954004,
-      "required_pages_per_second_upper_estimate": 0.3712545093566527
+      "required_token_positions_per_second": 29004.39173916682,
+      "required_pages_per_second_upper_estimate": 0.3720371947405346
     }
   ],
   "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": 0.20688801235158394,
-    "producer_p95_seconds": 0.1737280159995862,
-    "gpu_update_seconds": 0.9552445800726777,
-    "producer_batches_per_second": 4.833532830798371,
-    "gpu_batches_per_second": 1.046852314957827,
-    "producer_headroom": 4.6172060392234915,
+    "producer_batch_seconds": 0.07931229603120471,
+    "producer_p95_seconds": 0.16918425199946796,
+    "gpu_update_seconds": 0.9532349531283161,
+    "producer_batches_per_second": 12.608385459003216,
+    "gpu_batches_per_second": 1.0490593076955592,
+    "producer_headroom": 12.018753722036672,
     "sustainable": true,
-    "buffer_seconds": 60.18040854457869,
+    "buffer_seconds": 60.05380204708391,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 448790.0251071141,
-    "upload_bytes_per_second": 4680605.234342581,
+    "generation_bytes_per_second": 428386.4991190449,
+    "upload_bytes_per_second": 4765916.7777035665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 96.29989209089344,
+    "estimated_full_shards_per_hour": 91.9217703836299,
     "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": 1020572,
       "pages": 29,
       "shards": 1
+    },
+    {
+      "seconds": 1.2060352290000083,
+      "bytes": 11046742,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -55,3 +55,4 @@
 {"step": 402600, "loss": 0.5891031622886658, "lr": 0.0003, "grad_norm": 0.15020110006666054, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58930170858, "tokens_per_second": 153633.3075546813, "data_wait_s": 1.2636999599635601e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 402700, "loss": 0.5591861009597778, "lr": 0.0003, "grad_norm": 0.17413688342588102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58944884536, "tokens_per_second": 153173.0480065034, "data_wait_s": 1.420499938831199e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 402800, "loss": 0.5681113600730896, "lr": 0.0003, "grad_norm": 0.14310511750190794, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58959496916, "tokens_per_second": 152243.18869865793, "data_wait_s": 9.456000043428503e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 402900, "loss": 0.5551097989082336, "lr": 0.0003, "grad_norm": 0.16445451074476453, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 58973983599, "tokens_per_second": 150043.24327793208, "data_wait_s": 8.572999831812922e-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 sha256:4f2292de7228eee2f9537569293b26ba518f63aa6e7c2c6ff4a617e8d78f0dcb
-size 121674
+oid sha256:7d747bdc10a0ff0574cd4b5482a71d5c1d9f3ae2e3501a238ac96b6977e02003
+size 123792