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": 76249.2911685757,
-      "required_pages_per_second_upper_estimate": 0.005275102885772773
+      "required_token_positions_per_second": 75727.72480266062,
+      "required_pages_per_second_upper_estimate": 0.005239019714378066
     },
     {
       "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": 9531.161396071962,
-      "required_pages_per_second_upper_estimate": 0.003990498243252551
+      "required_token_positions_per_second": 9465.965600332578,
+      "required_pages_per_second_upper_estimate": 0.0039632021250195016
     },
     {
       "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": 19062.322792143925,
-      "required_pages_per_second_upper_estimate": 0.003970179130345952
+      "required_token_positions_per_second": 18931.931200665156,
+      "required_pages_per_second_upper_estimate": 0.003943022000498431
     },
     {
       "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": 28593.484188215887,
-      "required_pages_per_second_upper_estimate": 0.0017141357173611669
+      "required_token_positions_per_second": 28397.896800997732,
+      "required_pages_per_second_upper_estimate": 0.0017024105521420867
     },
     {
       "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": 28593.484188215887,
-      "required_pages_per_second_upper_estimate": 0.002223017626188566
+      "required_token_positions_per_second": 28397.896800997732,
+      "required_pages_per_second_upper_estimate": 0.002207811567130352
     },
     {
       "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": 28593.484188215887,
-      "required_pages_per_second_upper_estimate": 0.36676651387508996
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28397.896800997732,
+      "required_pages_per_second_upper_estimate": 0.36425772887723007
     }
   ],
   "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": 0.1140973576017359,
-    "producer_p95_seconds": 0.2705246689984051,
-    "gpu_update_seconds": 0.9669335789233567,
-    "producer_batches_per_second": 8.764444865503053,
-    "gpu_batches_per_second": 1.034197200094614,
-    "producer_headroom": 8.474636041077305,
+    "producer_batch_seconds": 0.21189846043753846,
+    "producer_p95_seconds": 0.28130581000004895,
+    "gpu_update_seconds": 0.9735932274755155,
+    "producer_batches_per_second": 4.7192414609108075,
+    "gpu_batches_per_second": 1.0271230035083092,
+    "producer_headroom": 4.59462152516442,
     "sustainable": true,
-    "buffer_seconds": 60.91681547217147,
+    "buffer_seconds": 60.36278010348196,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 172,
+    "upload_pages_per_commit": 166,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 598045.1742761765,
-    "upload_bytes_per_second": 3954040.5828353614,
+    "generation_bytes_per_second": 569951.374450408,
+    "upload_bytes_per_second": 3908127.104333609,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.16328631026254,
+    "estimated_full_shards_per_hour": 61.14914858405199,
     "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": 1.5756286089999776,
-      "bytes": 6163874,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.85335265699905,
       "bytes": 5996093,
@@ -562,6 +556,12 @@
       "bytes": 6263907,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.8874185050008236,
+      "bytes": 2469051,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -245,3 +245,4 @@
 {"step": 824500, "loss": 0.5239776968955994, "lr": 3e-05, "grad_norm": 0.274425546399414, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120683638925, "tokens_per_second": 150804.11986326615, "data_wait_s": 1.3046999811194837e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 824600, "loss": 0.5294896364212036, "lr": 3e-05, "grad_norm": 0.2926560772342064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120698312369, "tokens_per_second": 151075.42040384788, "data_wait_s": 1.2972999684279785e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 824700, "loss": 0.5112605690956116, "lr": 3e-05, "grad_norm": 0.2951030509979187, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120713050943, "tokens_per_second": 152679.64644456818, "data_wait_s": 1.1280000762781128e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 824800, "loss": 0.5212706923484802, "lr": 3e-05, "grad_norm": 0.2911333715258653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120727856599, "tokens_per_second": 151354.767689479, "data_wait_s": 9.175000741379336e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aa424d072ca24b9dc229a7019762afffeecaca30bc02914668e721bc093470bb
-size 526443
+oid sha256:b17c6984637469cc917b6a09c4c2bb2eaf7c481bde8bf4ab985a8e394e668a71
+size 528561