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": 76829.99701628789,
-      "required_pages_per_second_upper_estimate": 0.005315277463740708
+      "required_token_positions_per_second": 76533.89592404292,
+      "required_pages_per_second_upper_estimate": 0.00529479250313001
     },
     {
       "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": 9603.749627035986,
-      "required_pages_per_second_upper_estimate": 0.004020889419742556
+      "required_token_positions_per_second": 9566.736990505366,
+      "required_pages_per_second_upper_estimate": 0.004005393001738927
     },
     {
       "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": 19207.499254071972,
-      "required_pages_per_second_upper_estimate": 0.004000415558804804
+      "required_token_positions_per_second": 19133.47398101073,
+      "required_pages_per_second_upper_estimate": 0.003984998046603876
     },
     {
       "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": 28811.248881107957,
-      "required_pages_per_second_upper_estimate": 0.0017271903781926133
+      "required_token_positions_per_second": 28700.2109715161,
+      "required_pages_per_second_upper_estimate": 0.0017205338250576524
     },
     {
       "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": 28811.248881107957,
-      "required_pages_per_second_upper_estimate": 0.0022399478732152687
+      "required_token_positions_per_second": 28700.2109715161,
+      "required_pages_per_second_upper_estimate": 0.0022313151641486497
     },
     {
       "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": 28811.248881107957,
-      "required_pages_per_second_upper_estimate": 0.3695597655379992
+      "required_token_positions_per_second": 28700.2109715161,
+      "required_pages_per_second_upper_estimate": 0.36813549045697336
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10520952253901328,
-    "producer_p95_seconds": 0.2709512379988155,
-    "gpu_update_seconds": 0.959625183694459,
-    "producer_batches_per_second": 9.504843058566157,
-    "gpu_batches_per_second": 1.0420735272391477,
-    "producer_headroom": 9.121086766063552,
+    "producer_batch_seconds": 0.09261219503125062,
+    "producer_p95_seconds": 0.28489762600293034,
+    "gpu_update_seconds": 0.9633378663118407,
+    "producer_batches_per_second": 10.797714055503864,
+    "gpu_batches_per_second": 1.0380573991433775,
+    "producer_headroom": 10.401846819274466,
     "sustainable": true,
-    "buffer_seconds": 60.45638657275092,
+    "buffer_seconds": 60.690285577645966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 414877.5393981107,
-    "upload_bytes_per_second": 4974257.45299444,
+    "generation_bytes_per_second": 484206.1486468394,
+    "upload_bytes_per_second": 4969111.972692786,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.02306210000506,
+    "estimated_full_shards_per_hour": 103.89936775735748,
     "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": [
-    {
-      "seconds": 0.9664811299990106,
-      "bytes": 2719675,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.2011403209980926,
       "bytes": 11095830,
@@ -562,6 +556,12 @@
       "bytes": 6473769,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0211126419999346,
+      "bytes": 2557759,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -250,3 +250,4 @@
 {"step": 422100, "loss": 0.5418153405189514, "lr": 0.0003, "grad_norm": 0.16077626109792695, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61784167740, "tokens_per_second": 152017.29884691516, "data_wait_s": 1.5213998267427087e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 422200, "loss": 0.5677908658981323, "lr": 0.0003, "grad_norm": 0.14784672898380272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61798737904, "tokens_per_second": 150225.59618138464, "data_wait_s": 1.3166001735953614e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 422300, "loss": 0.5385948419570923, "lr": 0.0003, "grad_norm": 0.16267988688353516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61813394998, "tokens_per_second": 151461.4746913633, "data_wait_s": 1.1164000170538202e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 422400, "loss": 0.5615028142929077, "lr": 0.0003, "grad_norm": 0.18445800475886337, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61828069091, "tokens_per_second": 152599.17434878479, "data_wait_s": 2.7725000109057873e-05, "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:ec7a4670574e92b529c2a8d93d36a31055426cc2656e5da2b80c128e505829ab
-size 537816
+oid sha256:123fa1e9d653a3c6739550d107f4d857eb2e7eb59d7fc54aaa88eae52fb20e7a
+size 539934