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": 76701.57269713016,
-      "required_pages_per_second_upper_estimate": 0.005306392771355902
+      "required_token_positions_per_second": 76703.71848483548,
+      "required_pages_per_second_upper_estimate": 0.005306541221928269
     },
     {
       "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": 9587.69658714127,
-      "required_pages_per_second_upper_estimate": 0.004014168347164232
+      "required_token_positions_per_second": 9587.964810604435,
+      "required_pages_per_second_upper_estimate": 0.004014280646727117
     },
     {
       "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": 19175.39317428254,
-      "required_pages_per_second_upper_estimate": 0.003993728709078927
+      "required_token_positions_per_second": 19175.92962120887,
+      "required_pages_per_second_upper_estimate": 0.003993840436826625
     },
     {
       "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": 28763.089761423813,
-      "required_pages_per_second_upper_estimate": 0.0017243033125022668
+      "required_token_positions_per_second": 28763.894431813304,
+      "required_pages_per_second_upper_estimate": 0.0017243515512634562
     },
     {
       "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": 28763.089761423813,
-      "required_pages_per_second_upper_estimate": 0.0022362037134894073
+      "required_token_positions_per_second": 28763.894431813304,
+      "required_pages_per_second_upper_estimate": 0.002236266273072819
     },
     {
       "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": 28763.089761423813,
-      "required_pages_per_second_upper_estimate": 0.368942032059925
+      "required_token_positions_per_second": 28763.894431813304,
+      "required_pages_per_second_upper_estimate": 0.3689523535076936
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.08156788806249438,
-    "producer_p95_seconds": 0.2941333589988062,
-    "gpu_update_seconds": 0.9612319201214837,
-    "producer_batches_per_second": 12.259726514358642,
-    "gpu_batches_per_second": 1.0403316609311275,
-    "producer_headroom": 11.784440457561221,
+    "producer_batch_seconds": 0.12856291852338586,
+    "producer_p95_seconds": 0.32105255799979204,
+    "gpu_update_seconds": 0.9612050296437221,
+    "producer_batches_per_second": 7.778292617230044,
+    "gpu_batches_per_second": 1.0403607650395437,
+    "producer_headroom": 7.47653398572215,
     "sustainable": true,
-    "buffer_seconds": 60.557610967653474,
+    "buffer_seconds": 60.555916867554494,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 521219.3506287686,
-    "upload_bytes_per_second": 5180511.18667118,
+    "generation_bytes_per_second": 520231.5002343338,
+    "upload_bytes_per_second": 5115899.13953586,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.8415392794351,
+    "estimated_full_shards_per_hour": 111.62956958076964,
     "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": 1.1600354100010009,
-      "bytes": 11531600,
-      "pages": 28,
-      "shards": 3
-    },
     {
       "seconds": 1.4267704899975797,
       "bytes": 7513990,
@@ -562,6 +556,12 @@
       "bytes": 5851090,
       "pages": 31,
       "shards": 2
+    },
+    {
+      "seconds": 1.4330363720000605,
+      "bytes": 7577948,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -270,3 +270,4 @@
 {"step": 631400, "loss": 0.5551297068595886, "lr": 0.0001, "grad_norm": 0.1805686428261884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92420385230, "tokens_per_second": 151838.58818632044, "data_wait_s": 1.6391997633036226e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631500, "loss": 0.5649096369743347, "lr": 0.0001, "grad_norm": 0.17203733025526274, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92435093571, "tokens_per_second": 152675.37007872996, "data_wait_s": 1.105699993786402e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631600, "loss": 0.49820688366889954, "lr": 0.0001, "grad_norm": 0.16837591631471038, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92449576074, "tokens_per_second": 149642.83750909698, "data_wait_s": 1.789499947335571e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 631700, "loss": 0.5200865268707275, "lr": 0.0001, "grad_norm": 0.16134545844521925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92464076764, "tokens_per_second": 150018.86339453675, "data_wait_s": 1.6152000171132386e-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 sha256:91de5fa571aa5ef867cdc4240ba9d0e7754d536e0a7cf665f338c23191d73cc8
-size 580959
+oid sha256:9adb2d37b345ad1d45b3f19db324acda2fe995227023d5a476526ea78e84fe78
+size 583077