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": 75439.61911629433,
-      "required_pages_per_second_upper_estimate": 0.005219087894497953
+      "required_token_positions_per_second": 77349.25510783675,
+      "required_pages_per_second_upper_estimate": 0.005351200943358826
     },
     {
       "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": 9429.952389536791,
-      "required_pages_per_second_upper_estimate": 0.003948124145700664
+      "required_token_positions_per_second": 9668.656888479594,
+      "required_pages_per_second_upper_estimate": 0.004048064734691247
     },
     {
       "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": 18859.904779073582,
-      "required_pages_per_second_upper_estimate": 0.003928020796345378
+      "required_token_positions_per_second": 19337.31377695919,
+      "required_pages_per_second_upper_estimate": 0.004027452500483026
     },
     {
       "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": 28289.85716861037,
-      "required_pages_per_second_upper_estimate": 0.0016959337411474313
+      "required_token_positions_per_second": 29005.97066543878,
+      "required_pages_per_second_upper_estimate": 0.001738863651840297
     },
     {
       "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": 28289.85716861037,
-      "required_pages_per_second_upper_estimate": 0.0021994119609283548
+      "required_token_positions_per_second": 29005.97066543878,
+      "required_pages_per_second_upper_estimate": 0.0022550866354563775
     },
     {
       "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": 28289.85716861037,
-      "required_pages_per_second_upper_estimate": 0.3628719124768842
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29005.97066543878,
+      "required_pages_per_second_upper_estimate": 0.3720574475114324
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.07851594276552021,
-    "producer_p95_seconds": 0.19960272000025725,
-    "gpu_update_seconds": 0.9773114029956094,
-    "producer_batches_per_second": 12.736266862214178,
-    "gpu_batches_per_second": 1.0232153200452248,
-    "producer_headroom": 12.447298836037024,
+    "producer_batch_seconds": 0.05630373289831425,
+    "producer_p95_seconds": 0.1907463679999637,
+    "gpu_update_seconds": 0.9531830642352255,
+    "producer_batches_per_second": 17.760811735272,
+    "gpu_batches_per_second": 1.0491164158506503,
+    "producer_headroom": 16.929304953131517,
     "sustainable": true,
-    "buffer_seconds": 60.59330698572778,
+    "buffer_seconds": 60.050533046819204,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 87,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295407.8495669442,
-    "upload_bytes_per_second": 5467273.617418525,
+    "generation_bytes_per_second": 295401.8804208654,
+    "upload_bytes_per_second": 5413503.412403364,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.38764777427906,
+    "estimated_full_shards_per_hour": 63.38636693448516,
     "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": [
@@ -520,6 +520,12 @@
       "bytes": 7393031,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.8912479129994608,
+      "bytes": 1116344,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -104,3 +104,4 @@
 {"step": 770400, "loss": 0.5213112831115723, "lr": 3e-05, "grad_norm": 0.23726087482793934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112764078349, "tokens_per_second": 153326.7860213534, "data_wait_s": 1.5965999409672804e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 770500, "loss": 0.5744532942771912, "lr": 3e-05, "grad_norm": 0.24345761887305634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112778741029, "tokens_per_second": 151479.51465090388, "data_wait_s": 7.807000656612217e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 770600, "loss": 0.5072928071022034, "lr": 3e-05, "grad_norm": 0.25135627223391616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112793444564, "tokens_per_second": 153555.77624449757, "data_wait_s": 1.302700002270285e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 770700, "loss": 0.6074251532554626, "lr": 3e-05, "grad_norm": 0.2844986677756334, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 112808199028, "tokens_per_second": 153507.88402762378, "data_wait_s": 1.2989999959245324e-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 sha256:5a9cd0255d15d60fa2a11935a4ba20ec2c016164eb823d6f688ae4e75efb7877
-size 226239
+oid sha256:964cf99ef97c404d8264198a00006c448f76d971479b647185eb414c0d8c915c
+size 228357