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": 76960.20840364542,
-      "required_pages_per_second_upper_estimate": 0.005324285789650146
+      "required_token_positions_per_second": 77196.27955109101,
+      "required_pages_per_second_upper_estimate": 0.0053406177393909605
     },
     {
       "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": 9620.026050455677,
-      "required_pages_per_second_upper_estimate": 0.00402770401833801
+      "required_token_positions_per_second": 9649.534943886376,
+      "required_pages_per_second_upper_estimate": 0.004040058775801676
     },
     {
       "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": 19240.052100911355,
-      "required_pages_per_second_upper_estimate": 0.004007195458325146
+      "required_token_positions_per_second": 19299.069887772752,
+      "required_pages_per_second_upper_estimate": 0.004019487306924672
     },
     {
       "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": 28860.078151367034,
-      "required_pages_per_second_upper_estimate": 0.0017301176183866663
+      "required_token_positions_per_second": 28948.604831659126,
+      "required_pages_per_second_upper_estimate": 0.0017354246576977632
     },
     {
       "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": 28860.078151367034,
-      "required_pages_per_second_upper_estimate": 0.0022437441342006504
+      "required_token_positions_per_second": 28948.604831659126,
+      "required_pages_per_second_upper_estimate": 0.0022506266942056456
     },
     {
       "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": 28860.078151367034,
-      "required_pages_per_second_upper_estimate": 0.37018609498809707
+      "required_token_positions_per_second": 28948.604831659126,
+      "required_pages_per_second_upper_estimate": 0.3713216201903404
     }
   ],
   "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": 0.2625593772421553,
-    "producer_p95_seconds": 0.24768189499809523,
-    "gpu_update_seconds": 0.9580015637861459,
-    "producer_batches_per_second": 3.8086622938540575,
-    "gpu_batches_per_second": 1.0438396322109025,
-    "producer_headroom": 3.6487044334455168,
+    "producer_batch_seconds": 0.08863720699201849,
+    "producer_p95_seconds": 0.24648081300256308,
+    "gpu_update_seconds": 0.9550719338903427,
+    "producer_batches_per_second": 11.28194393681704,
+    "gpu_batches_per_second": 1.0470415520710041,
+    "producer_headroom": 10.775068013778277,
     "sustainable": true,
-    "buffer_seconds": 60.35409851852719,
+    "buffer_seconds": 60.16953183509159,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342426.661330425,
-    "upload_bytes_per_second": 5030840.28034647,
+    "generation_bytes_per_second": 343376.9829683435,
+    "upload_bytes_per_second": 5118760.807672537,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.4767902368027,
+    "estimated_full_shards_per_hour": 73.6807071379445,
     "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": 1.041803035999692,
-      "bytes": 1241755,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.3904252129996166,
       "bytes": 12223668,
@@ -562,6 +556,12 @@
       "bytes": 7320475,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.333093295001163,
+      "bytes": 10300175,
+      "pages": 35,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -237,3 +237,4 @@
 {"step": 783700, "loss": 0.5481111407279968, "lr": 3e-05, "grad_norm": 0.26958007243315957, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114711391951, "tokens_per_second": 140675.79402215322, "data_wait_s": 1.4942001143936068e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783800, "loss": 0.5824326276779175, "lr": 3e-05, "grad_norm": 0.2746221799751516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114726173008, "tokens_per_second": 153272.93106106776, "data_wait_s": 1.0961997759295627e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 783900, "loss": 0.5601587891578674, "lr": 3e-05, "grad_norm": 0.2715992571080736, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114740873998, "tokens_per_second": 153443.8424656105, "data_wait_s": 8.216000423999503e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 784000, "loss": 0.5442437529563904, "lr": 3e-05, "grad_norm": 0.26176771027942847, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114755616200, "tokens_per_second": 152848.74576709841, "data_wait_s": 1.0533000022405759e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c038f4271c8f95b1ecb79b64624aa07eba600b9a99d9f78d942385f3e02bc33f
-size 509499
+oid sha256:045831b3606d947f68daf4d5190d11cde0ae39373c5ae26cf05610a15c5219e2
+size 511617