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": 76794.99568323781,
-      "required_pages_per_second_upper_estimate": 0.005312855990306022
+      "required_token_positions_per_second": 76916.98852569632,
+      "required_pages_per_second_upper_estimate": 0.005321295738209694
     },
     {
       "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": 9599.374460404726,
-      "required_pages_per_second_upper_estimate": 0.004019057628837917
+      "required_token_positions_per_second": 9614.62356571204,
+      "required_pages_per_second_upper_estimate": 0.00402544211079256
     },
     {
       "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": 19198.748920809452,
-      "required_pages_per_second_upper_estimate": 0.0039985930951480265
+      "required_token_positions_per_second": 19229.24713142408,
+      "required_pages_per_second_upper_estimate": 0.004004945068127153
     },
     {
       "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": 28798.12338121418,
-      "required_pages_per_second_upper_estimate": 0.0017264035245154593
+      "required_token_positions_per_second": 28843.87069713612,
+      "required_pages_per_second_upper_estimate": 0.0017291460062528728
     },
     {
       "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": 28798.12338121418,
-      "required_pages_per_second_upper_estimate": 0.0022389274233315
+      "required_token_positions_per_second": 28843.87069713612,
+      "required_pages_per_second_upper_estimate": 0.002242484075923254
     },
     {
       "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": 28798.12338121418,
-      "required_pages_per_second_upper_estimate": 0.3693914057184256
+      "required_token_positions_per_second": 28843.87069713612,
+      "required_pages_per_second_upper_estimate": 0.3699782031674314
     }
   ],
   "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": 0.09641828085167958,
-    "producer_p95_seconds": 0.24993470199842704,
-    "gpu_update_seconds": 0.9600625580358324,
-    "producer_batches_per_second": 10.37147718427278,
-    "gpu_batches_per_second": 1.0415987912765545,
-    "producer_headroom": 9.9572669161432,
+    "producer_batch_seconds": 0.07018635236713067,
+    "producer_p95_seconds": 0.26296824800010654,
+    "gpu_update_seconds": 0.9585398676310507,
+    "producer_batches_per_second": 14.247784167058597,
+    "gpu_batches_per_second": 1.0432534250989627,
+    "producer_headroom": 13.657069149528128,
     "sustainable": true,
-    "buffer_seconds": 60.483941156257444,
+    "buffer_seconds": 60.38801166075619,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 363115.05823380576,
+    "generation_bytes_per_second": 352319.3076511126,
     "upload_bytes_per_second": 4998268.099658367,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 77.91603861103658,
+    "estimated_full_shards_per_hour": 75.59952184820207,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -219,3 +219,4 @@
 {"step": 419000, "loss": 0.6161655187606812, "lr": 0.0003, "grad_norm": 0.17562282216041825, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61330665018, "tokens_per_second": 152111.1166580027, "data_wait_s": 1.2300002708798274e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 419100, "loss": 0.5930498838424683, "lr": 0.0003, "grad_norm": 0.17148425860614736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61345397171, "tokens_per_second": 152520.6746918214, "data_wait_s": 1.1173000530106947e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 419200, "loss": 0.5675833821296692, "lr": 0.0003, "grad_norm": 0.1525742805133259, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61359906081, "tokens_per_second": 150730.13709377113, "data_wait_s": 1.671099744271487e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 419300, "loss": 0.6181345582008362, "lr": 0.0003, "grad_norm": 0.1895822870259062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61374469007, "tokens_per_second": 151222.4293424605, "data_wait_s": 1.4278000890044495e-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 sha256:8314463a321066b47e263bd3d872b444ac633796b58f843234be78013481beae
-size 471375
+oid sha256:c02ea2437164f8ba756cf1cfe4553be215a783bb1b9c941b7887f48b3f5aa004
+size 473493