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": 77150.8938292305,
-      "required_pages_per_second_upper_estimate": 0.005337477849843267
+      "required_token_positions_per_second": 77119.4986456471,
+      "required_pages_per_second_upper_estimate": 0.005335305858196106
     },
     {
       "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": 9643.861728653812,
-      "required_pages_per_second_upper_estimate": 0.0040376835190540075
+      "required_token_positions_per_second": 9639.937330705887,
+      "required_pages_per_second_upper_estimate": 0.00403604045558396
     },
     {
       "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": 19287.723457307624,
-      "required_pages_per_second_upper_estimate": 0.004017124144684279
+      "required_token_positions_per_second": 19279.874661411774,
+      "required_pages_per_second_upper_estimate": 0.004015489447485841
     },
     {
       "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": 28931.585185961438,
-      "required_pages_per_second_upper_estimate": 0.001734404355925681
+      "required_token_positions_per_second": 28919.81199211766,
+      "required_pages_per_second_upper_estimate": 0.0017336985709313726
     },
     {
       "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": 28931.585185961438,
-      "required_pages_per_second_upper_estimate": 0.00224930349161416
+      "required_token_positions_per_second": 28919.81199211766,
+      "required_pages_per_second_upper_estimate": 0.0022483881775776132
     },
     {
       "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": 28931.585185961438,
-      "required_pages_per_second_upper_estimate": 0.3711033104496022
+      "required_token_positions_per_second": 28919.81199211766,
+      "required_pages_per_second_upper_estimate": 0.37095229656004486
     }
   ],
   "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": 0.08143478721109432,
-    "producer_p95_seconds": 0.21723602999918512,
-    "gpu_update_seconds": 0.955633776106251,
-    "producer_batches_per_second": 12.279764388747177,
-    "gpu_batches_per_second": 1.0464259688209432,
-    "producer_headroom": 11.734957612513533,
+    "producer_batch_seconds": 0.1069210857264693,
+    "producer_p95_seconds": 0.23104409200095688,
+    "gpu_update_seconds": 0.9560228125803756,
+    "producer_batches_per_second": 9.35269215801127,
+    "gpu_batches_per_second": 1.0460001443908298,
+    "producer_headroom": 8.941387062100356,
     "sustainable": true,
-    "buffer_seconds": 60.20492789469381,
+    "buffer_seconds": 60.22943719256366,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 461542.5028301368,
+    "generation_bytes_per_second": 484782.9837068155,
     "upload_bytes_per_second": 5344283.020856207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 99.0362769477661,
+    "estimated_full_shards_per_hour": 104.02314313319539,
     "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

@@ -172,3 +172,4 @@
 {"step": 621600, "loss": 0.5672624707221985, "lr": 0.0001, "grad_norm": 0.15244863120421648, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 90985525084, "tokens_per_second": 151131.1363444428, "data_wait_s": 1.2983997294213623e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 621700, "loss": 0.49639052152633667, "lr": 0.0001, "grad_norm": 0.17698319896955775, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91000274262, "tokens_per_second": 153517.07150469164, "data_wait_s": 1.2032000086037442e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 621800, "loss": 0.5697305202484131, "lr": 0.0001, "grad_norm": 0.15693798460805822, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91014969989, "tokens_per_second": 153217.47249473794, "data_wait_s": 1.4692999684484676e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 621900, "loss": 0.5458886027336121, "lr": 0.0001, "grad_norm": 0.16189777602555494, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 91029653225, "tokens_per_second": 152145.4358169142, "data_wait_s": 1.0513998859096318e-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 sha256:25e7ec2fdedab2c37247eff89463e7a3bf67d09d9486cbc45059684ad803c09c
-size 371829
+oid sha256:116a4eba7b0630a4b84e61008b0383e9ed32a299f8cbf2f32c92074e4e8dda1e
+size 373947