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": 76866.19861380632,
-      "required_pages_per_second_upper_estimate": 0.0053177819742563125
+      "required_token_positions_per_second": 76730.50169681945,
+      "required_pages_per_second_upper_estimate": 0.005308394146678934
     },
     {
       "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": 9608.27482672579,
-      "required_pages_per_second_upper_estimate": 0.004022784026355763
+      "required_token_positions_per_second": 9591.312712102432,
+      "required_pages_per_second_upper_estimate": 0.004015682343172194
     },
     {
       "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": 19216.54965345158,
-      "required_pages_per_second_upper_estimate": 0.0040023005183204945
+      "required_token_positions_per_second": 19182.625424204864,
+      "required_pages_per_second_upper_estimate": 0.0039952349960104905
     },
     {
       "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": 28824.824480177373,
-      "required_pages_per_second_upper_estimate": 0.0017280042146280853
+      "required_token_positions_per_second": 28773.938136307297,
+      "required_pages_per_second_upper_estimate": 0.0017249536560120225
     },
     {
       "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": 28824.824480177373,
-      "required_pages_per_second_upper_estimate": 0.002241003316330166
+      "required_token_positions_per_second": 28773.938136307297,
+      "required_pages_per_second_upper_estimate": 0.0022370471269196395
     },
     {
       "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": 28824.824480177373,
-      "required_pages_per_second_upper_estimate": 0.3697338987465191
+      "required_token_positions_per_second": 28773.938136307297,
+      "required_pages_per_second_upper_estimate": 0.3690811833648529
     }
   ],
   "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": 0.3537641362970021,
-    "producer_p95_seconds": 0.3002260319990455,
-    "gpu_update_seconds": 0.9591732299710388,
-    "producer_batches_per_second": 2.8267421634861587,
-    "gpu_batches_per_second": 1.0425645428304895,
-    "producer_headroom": 2.711335411246341,
+    "producer_batch_seconds": 0.09437909564067581,
+    "producer_p95_seconds": 0.2617566420012736,
+    "gpu_update_seconds": 0.9608695156369098,
+    "producer_batches_per_second": 10.595566668780588,
+    "gpu_batches_per_second": 1.0407240356013925,
+    "producer_headroom": 10.18095701292979,
     "sustainable": true,
-    "buffer_seconds": 60.42791348817545,
+    "buffer_seconds": 60.53477948512532,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445869.368056448,
-    "upload_bytes_per_second": 4905965.722400359,
+    "generation_bytes_per_second": 462726.8381495783,
+    "upload_bytes_per_second": 5026241.350261635,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.67318707723693,
+    "estimated_full_shards_per_hour": 99.29040773740303,
     "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": 1.6562018110016652,
-      "bytes": 1121250,
-      "pages": 33,
-      "shards": 1
-    },
     {
       "seconds": 0.9384285160012951,
       "bytes": 5824612,
@@ -562,6 +556,12 @@
       "bytes": 10317954,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.058006204002595,
+      "bytes": 8327609,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -254,3 +254,4 @@
 {"step": 422500, "loss": 0.5777281522750854, "lr": 0.0003, "grad_norm": 0.15544418884510952, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61842752733, "tokens_per_second": 152067.14642873252, "data_wait_s": 1.0724001185735688e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 422600, "loss": 0.5908933281898499, "lr": 0.0003, "grad_norm": 0.17573596160658406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61857382235, "tokens_per_second": 150614.95424845358, "data_wait_s": 1.8890998035203665e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 422700, "loss": 0.5855945348739624, "lr": 0.0003, "grad_norm": 0.15746786691532302, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61871966723, "tokens_per_second": 150422.87907787634, "data_wait_s": 1.0731997463153675e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 422800, "loss": 0.5643939971923828, "lr": 0.0003, "grad_norm": 0.18316346773762682, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61886690249, "tokens_per_second": 152335.3523300517, "data_wait_s": 9.429000783711672e-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 sha256:6b2760c5f93849202654db68bdba3e0c11a68af87c458e3c685cd2c0c14574f6
-size 546288
+oid sha256:a3b935461bef2d7b406fae12ce3f82e47ceb586b95b76924e0b48a06c520d1a5
+size 548406