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": 75334.75334039006,
-      "required_pages_per_second_upper_estimate": 0.005211833036798772
+      "required_token_positions_per_second": 76807.15886398792,
+      "required_pages_per_second_upper_estimate": 0.005313697467372782
     },
     {
       "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": 9416.844167548757,
-      "required_pages_per_second_upper_estimate": 0.003942636006885077
+      "required_token_positions_per_second": 9600.89485799849,
+      "required_pages_per_second_upper_estimate": 0.004019694187560914
     },
     {
       "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": 18833.688335097515,
-      "required_pages_per_second_upper_estimate": 0.003922560602439283
+      "required_token_positions_per_second": 19201.78971599698,
+      "required_pages_per_second_upper_estimate": 0.003999226412594427
     },
     {
       "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": 28250.532502646274,
-      "required_pages_per_second_upper_estimate": 0.001693576287468172
+      "required_token_positions_per_second": 28802.68457399547,
+      "required_pages_per_second_upper_estimate": 0.0017266769610581594
     },
     {
       "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": 28250.532502646274,
-      "required_pages_per_second_upper_estimate": 0.0021963546411205718
+      "required_token_positions_per_second": 28802.68457399547,
+      "required_pages_per_second_upper_estimate": 0.0022392820360076786
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28250.532502646274,
-      "required_pages_per_second_upper_estimate": 0.36236749788543343
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28802.68457399547,
+      "required_pages_per_second_upper_estimate": 0.36944991180199677
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.1048541820704827,
-    "producer_p95_seconds": 0.3682760690062423,
-    "gpu_update_seconds": 0.978671817864324,
-    "producer_batches_per_second": 9.537054033074263,
-    "gpu_batches_per_second": 1.0217929869302038,
-    "producer_headroom": 9.333646007619071,
+    "producer_batch_seconds": 0.16223753505403238,
+    "producer_p95_seconds": 0.37372767699707765,
+    "gpu_update_seconds": 0.9599105225407364,
+    "producer_batches_per_second": 6.163801734703101,
+    "gpu_batches_per_second": 1.0417637649737945,
+    "producer_headroom": 5.916698143996351,
     "sustainable": true,
-    "buffer_seconds": 60.67765270758809,
+    "buffer_seconds": 60.47436292006639,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 653417.8932605253,
-    "upload_bytes_per_second": 5038073.572900272,
+    "generation_bytes_per_second": 641322.9943359176,
+    "upload_bytes_per_second": 4982639.792340205,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.1041345518199,
+    "estimated_full_shards_per_hour": 68.806492674628,
     "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": 1.1513834180004778,
-      "bytes": 6077960,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.0119541929961997,
       "bytes": 2485722,
@@ -562,6 +556,12 @@
       "bytes": 15819770,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0710374850023072,
+      "bytes": 956267,
+      "pages": 29,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -425,3 +425,4 @@
 {"step": 967500, "loss": 0.5437011122703552, "lr": 3e-05, "grad_norm": 0.3618753728928093, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141615881872, "tokens_per_second": 151832.95373416174, "data_wait_s": 1.6120000509545207e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 967600, "loss": 0.5181931257247925, "lr": 3e-05, "grad_norm": 0.3292315335811378, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141630607289, "tokens_per_second": 152958.61733542618, "data_wait_s": 8.55499820318073e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 967700, "loss": 0.507817804813385, "lr": 3e-05, "grad_norm": 0.34821866907319726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141645206040, "tokens_per_second": 151648.98401438427, "data_wait_s": 1.2882002920378e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 967800, "loss": 0.5485464930534363, "lr": 3e-05, "grad_norm": 0.3445373450443752, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 141659705637, "tokens_per_second": 150332.2215349967, "data_wait_s": 1.1493997590150684e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0694d7410699f572bc62d4080786bd54dcae850cd1cacdcc5c46c2fd7edd193d
-size 910815
+oid sha256:09eff846be7e7b4c49c80eac48a58e1fc1e5875e11056f95cdb1db1a0d8e6e60
+size 912933