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": 76904.63477392794,
-      "required_pages_per_second_upper_estimate": 0.005320441076997705
+      "required_token_positions_per_second": 76312.57664730771,
+      "required_pages_per_second_upper_estimate": 0.005279481122033996
     },
     {
       "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": 9613.079346740993,
-      "required_pages_per_second_upper_estimate": 0.004024795578556341
+      "required_token_positions_per_second": 9539.072080913464,
+      "required_pages_per_second_upper_estimate": 0.003993810281801804
     },
     {
       "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": 19226.158693481986,
-      "required_pages_per_second_upper_estimate": 0.004004301827951401
+      "required_token_positions_per_second": 19078.144161826927,
+      "required_pages_per_second_upper_estimate": 0.003973474304413345
     },
     {
       "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": 28839.23804022298,
-      "required_pages_per_second_upper_estimate": 0.0017288682855445875
+      "required_token_positions_per_second": 28617.21624274039,
+      "required_pages_per_second_upper_estimate": 0.0017155584177931617
     },
     {
       "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": 28839.23804022298,
-      "required_pages_per_second_upper_estimate": 0.0022421239072251613
+      "required_token_positions_per_second": 28617.21624274039,
+      "required_pages_per_second_upper_estimate": 0.00222486268904157
     },
     {
       "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": 28839.23804022298,
-      "required_pages_per_second_upper_estimate": 0.36991878041870907
+      "required_token_positions_per_second": 28617.21624274039,
+      "required_pages_per_second_upper_estimate": 0.3670709231890354
     }
   ],
   "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": 0.3065585571328526,
+    "producer_batch_seconds": 0.4560349880312913,
     "producer_p95_seconds": 0.17642489700028818,
-    "gpu_update_seconds": 0.9586938448733797,
-    "producer_batches_per_second": 3.2620195285125644,
-    "gpu_batches_per_second": 1.0430858666168612,
-    "producer_headroom": 3.12727804384176,
+    "gpu_update_seconds": 0.966131707762237,
+    "producer_batches_per_second": 2.1928142055876294,
+    "gpu_batches_per_second": 1.0350555643352282,
+    "producer_headroom": 2.118547333249669,
     "sustainable": true,
-    "buffer_seconds": 60.39771222702292,
+    "buffer_seconds": 60.86629758902093,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 334348.6825692587,
-    "upload_bytes_per_second": 5085743.629061496,
+    "generation_bytes_per_second": 308171.58652610425,
+    "upload_bytes_per_second": 5111276.780634591,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.74344404037781,
+    "estimated_full_shards_per_hour": 66.1264486011252,
     "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": [
@@ -364,6 +364,12 @@
       "bytes": 6046442,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1579751299996133,
+      "bytes": 6933490,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -50,3 +50,4 @@
 {"step": 890000, "loss": 0.5526643991470337, "lr": 3e-05, "grad_norm": 0.32601227166011965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130271344653, "tokens_per_second": 152680.7386107045, "data_wait_s": 1.1977999747614376e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 890100, "loss": 0.4908837080001831, "lr": 3e-05, "grad_norm": 0.3075500939918485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130285879941, "tokens_per_second": 139867.4475602635, "data_wait_s": 9.98499945126241e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 890200, "loss": 0.5247633457183838, "lr": 3e-05, "grad_norm": 0.3070394859758585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130300368156, "tokens_per_second": 150711.66640688496, "data_wait_s": 1.783500010787975e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 890300, "loss": 0.5213205218315125, "lr": 3e-05, "grad_norm": 0.3233589255494616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130315114383, "tokens_per_second": 154146.30505007337, "data_wait_s": 7.2249995355377905e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9125ee204cef82afce18f274cfe1a8f4af30db9310022c577bfc1690bf7564a0
-size 111084
+oid sha256:ff60d812f96d37cc496c71f965c31a527b6f0b027882f3a5abe7ba61f24d357d
+size 113202