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": 77487.92181529848,
-      "required_pages_per_second_upper_estimate": 0.005360794227932119
+      "required_token_positions_per_second": 75361.55902436309,
+      "required_pages_per_second_upper_estimate": 0.0052136875162137825
     },
     {
       "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": 9685.99022691231,
-      "required_pages_per_second_upper_estimate": 0.004055321841531759
+      "required_token_positions_per_second": 9420.194878045386,
+      "required_pages_per_second_upper_estimate": 0.003944038879399223
     },
     {
       "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": 19371.98045382462,
-      "required_pages_per_second_upper_estimate": 0.004034672655052348
+      "required_token_positions_per_second": 18840.38975609077,
+      "required_pages_per_second_upper_estimate": 0.0039239563317038225
     },
     {
       "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": 29057.97068073693,
-      "required_pages_per_second_upper_estimate": 0.001741980973357995
+      "required_token_positions_per_second": 28260.584634136158,
+      "required_pages_per_second_upper_estimate": 0.0016941788974022742
     },
     {
       "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": 29057.97068073693,
-      "required_pages_per_second_upper_estimate": 0.0022591294079219136
+      "required_token_positions_per_second": 28260.584634136158,
+      "required_pages_per_second_upper_estimate": 0.0021971361501292564
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 29057.97068073693,
-      "required_pages_per_second_upper_estimate": 0.37272444787441067
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28260.584634136158,
+      "required_pages_per_second_upper_estimate": 0.36249643583504776
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10769445542211997,
-    "producer_p95_seconds": 0.3145365979944472,
-    "gpu_update_seconds": 0.9514773176617035,
-    "producer_batches_per_second": 9.28552910250015,
-    "gpu_batches_per_second": 1.050997203441006,
-    "producer_headroom": 8.834970323516528,
+    "producer_batch_seconds": 0.11930168000014874,
+    "producer_p95_seconds": 0.3206528500013519,
+    "gpu_update_seconds": 0.9783237097863782,
+    "producer_batches_per_second": 8.382111634964012,
+    "gpu_batches_per_second": 1.0221565622879107,
+    "producer_headroom": 8.200418550561555,
     "sustainable": true,
-    "buffer_seconds": 60.894548330349025,
+    "buffer_seconds": 60.65607000675545,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 367140.17044661933,
-    "upload_bytes_per_second": 5161192.984974496,
+    "generation_bytes_per_second": 341924.23570351995,
+    "upload_bytes_per_second": 5143785.90882166,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.7797339920896,
+    "estimated_full_shards_per_hour": 73.36898139313888,
     "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": 0.9699245899973903,
-      "bytes": 5017188,
-      "pages": 6,
-      "shards": 2
-    },
     {
       "seconds": 0.8813080820000323,
       "bytes": 1258193,
@@ -562,6 +556,12 @@
       "bytes": 6736460,
       "pages": 41,
       "shards": 4
+    },
+    {
+      "seconds": 1.445422744996904,
+      "bytes": 6026824,
+      "pages": 21,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -334,3 +334,4 @@
 {"step": 637800, "loss": 0.5254839062690735, "lr": 0.0001, "grad_norm": 0.1608861134915749, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93357526174, "tokens_per_second": 152757.9456056643, "data_wait_s": 1.2194002920296043e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 637900, "loss": 0.5674863457679749, "lr": 0.0001, "grad_norm": 0.16188142427010585, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93372038230, "tokens_per_second": 150246.5437670056, "data_wait_s": 1.0493000445421785e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 638000, "loss": 0.5121858716011047, "lr": 0.0001, "grad_norm": 0.18508731141413415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93386675116, "tokens_per_second": 143104.28380744057, "data_wait_s": 1.0420000762678683e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 638100, "loss": 0.5641270279884338, "lr": 0.0001, "grad_norm": 0.20221628744098225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 93401303482, "tokens_per_second": 152214.43562992098, "data_wait_s": 1.549000444356352e-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 sha256:7f2b5d6013d52994e5a335f12d7822780b9c9a265741b8877acf83dfff5290dc
-size 717294
+oid sha256:83e8101a85788bb0c06a08ab885bf604882ca3a8f7fc1d49927f7bf5851ffe8f
+size 719412