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": 76281.05800507293,
-      "required_pages_per_second_upper_estimate": 0.005277300589230868
+      "required_token_positions_per_second": 76160.83238538008,
+      "required_pages_per_second_upper_estimate": 0.005268983101898645
     },
     {
       "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": 9535.132250634117,
-      "required_pages_per_second_upper_estimate": 0.003992160757136853
+      "required_token_positions_per_second": 9520.10404817251,
+      "required_pages_per_second_upper_estimate": 0.003985868762590732
     },
     {
       "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": 19070.264501268233,
-      "required_pages_per_second_upper_estimate": 0.003971833178919592
+      "required_token_positions_per_second": 19040.20809634502,
+      "required_pages_per_second_upper_estimate": 0.003965573222414787
     },
     {
       "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": 28605.396751902346,
-      "required_pages_per_second_upper_estimate": 0.0017148498573647387
+      "required_token_positions_per_second": 28560.31214451753,
+      "required_pages_per_second_upper_estimate": 0.0017121471039922285
     },
     {
       "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": 28605.396751902346,
-      "required_pages_per_second_upper_estimate": 0.002223943775617358
+      "required_token_positions_per_second": 28560.31214451753,
+      "required_pages_per_second_upper_estimate": 0.0022204386456994188
     },
     {
       "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": 28605.396751902346,
-      "required_pages_per_second_upper_estimate": 0.3669193154513455
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28560.31214451753,
+      "required_pages_per_second_upper_estimate": 0.36634101851589296
     }
   ],
   "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": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10162633339851368,
-    "producer_p95_seconds": 0.2952909859995998,
-    "gpu_update_seconds": 0.9665309046329281,
-    "producer_batches_per_second": 9.8399692929847,
-    "gpu_batches_per_second": 1.0346280653899866,
-    "producer_headroom": 9.510634422308735,
+    "producer_batch_seconds": 0.07803326889066398,
+    "producer_p95_seconds": 0.28913194900087547,
+    "gpu_update_seconds": 0.9680566465835122,
+    "producer_batches_per_second": 12.815046892385686,
+    "gpu_batches_per_second": 1.0329974010603853,
+    "producer_headroom": 12.405691320453347,
     "sustainable": true,
-    "buffer_seconds": 60.891446991874474,
+    "buffer_seconds": 60.019512088177756,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 88,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 297148.27298900066,
-    "upload_bytes_per_second": 4709981.621861987,
+    "generation_bytes_per_second": 296431.60070001805,
+    "upload_bytes_per_second": 4776004.179462634,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.761102125668664,
+    "estimated_full_shards_per_hour": 63.60732093573004,
     "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": 1.2758062670000072,
-      "bytes": 1083353,
-      "pages": 30,
-      "shards": 1
-    },
     {
       "seconds": 1.7504430440021679,
       "bytes": 10417035,
@@ -562,6 +556,12 @@
       "bytes": 1138613,
       "pages": 35,
       "shards": 1
+    },
+    {
+      "seconds": 1.0960563930020726,
+      "bytes": 6111629,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -263,3 +263,4 @@
 {"step": 868800, "loss": 0.5246954560279846, "lr": 3e-05, "grad_norm": 0.3131322136905751, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127168232000, "tokens_per_second": 151019.21597509962, "data_wait_s": 9.941999451257288e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 868900, "loss": 0.524967610836029, "lr": 3e-05, "grad_norm": 0.3019740572453568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127182912800, "tokens_per_second": 152499.50259307757, "data_wait_s": 1.1255000572418794e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 869000, "loss": 0.49112239480018616, "lr": 3e-05, "grad_norm": 0.29584658459382734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127197639142, "tokens_per_second": 153049.0122783494, "data_wait_s": 1.383099879603833e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 869100, "loss": 0.5332478880882263, "lr": 3e-05, "grad_norm": 0.3049947083564535, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 127212320735, "tokens_per_second": 151972.2351776996, "data_wait_s": 1.2565997167257592e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:685e191446d8e319563a70ca81981c56d0063b9d683df820adfc12460dba1449
-size 565350
+oid sha256:31e776c9b9c88d893126ce13f52d1a60cfc53745315170cf5c9f50bdfb532f70
+size 567468