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": 76341.49263198496,
-      "required_pages_per_second_upper_estimate": 0.005281481596948296
+      "required_token_positions_per_second": 76724.38363244286,
+      "required_pages_per_second_upper_estimate": 0.0053079708847894955
     },
     {
       "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": 9542.68657899812,
-      "required_pages_per_second_upper_estimate": 0.003995323596670547
+      "required_token_positions_per_second": 9590.547954055357,
+      "required_pages_per_second_upper_estimate": 0.004015362154947848
     },
     {
       "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": 19085.37315799624,
-      "required_pages_per_second_upper_estimate": 0.003974979913673964
+      "required_token_positions_per_second": 19181.095908110714,
+      "required_pages_per_second_upper_estimate": 0.003994916438144131
     },
     {
       "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": 28628.05973699436,
-      "required_pages_per_second_upper_estimate": 0.0017162084687166303
+      "required_token_positions_per_second": 28771.643862166075,
+      "required_pages_per_second_upper_estimate": 0.0017248161177804099
     },
     {
       "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": 28628.05973699436,
-      "required_pages_per_second_upper_estimate": 0.002225705723024326
+      "required_token_positions_per_second": 28771.643862166075,
+      "required_pages_per_second_upper_estimate": 0.0022368687571966053
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28628.05973699436,
-      "required_pages_per_second_upper_estimate": 0.36721001189048835
+      "required_token_positions_per_second": 28771.643862166075,
+      "required_pages_per_second_upper_estimate": 0.3690517548795689
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.10519405938262594,
-    "producer_p95_seconds": 0.3276846029984881,
-    "gpu_update_seconds": 0.965765764568114,
-    "producer_batches_per_second": 9.506240237033404,
-    "gpu_batches_per_second": 1.0354477624780947,
-    "producer_headroom": 9.180801370686735,
+    "producer_batch_seconds": 0.16628991921862735,
+    "producer_p95_seconds": 0.3130666650031344,
+    "gpu_update_seconds": 0.9609461361488756,
+    "producer_batches_per_second": 6.013593636336211,
+    "gpu_batches_per_second": 1.040641054042465,
+    "producer_headroom": 5.778739569206748,
     "sustainable": true,
-    "buffer_seconds": 60.84324316779119,
+    "buffer_seconds": 60.53960657737916,
     "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": 454614.292029797,
-    "upload_bytes_per_second": 6116695.635149223,
+    "generation_bytes_per_second": 459354.33187024965,
+    "upload_bytes_per_second": 6082411.288184336,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.54964419050629,
+    "estimated_full_shards_per_hour": 98.56674639778726,
     "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": 1.1295750149984087,
-      "bytes": 5944567,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.1342636209992634,
       "bytes": 6342296,
@@ -562,6 +556,12 @@
       "bytes": 6070118,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9898074049997376,
+      "bytes": 2654933,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -292,3 +292,4 @@
 {"step": 514200, "loss": 0.5733696222305298, "lr": 0.0003, "grad_norm": 0.14613767507801634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75261727650, "tokens_per_second": 152225.59964747558, "data_wait_s": 1.2679000064963475e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 514300, "loss": 0.5835843682289124, "lr": 0.0003, "grad_norm": 0.17735822378102903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75276183232, "tokens_per_second": 150682.23349223693, "data_wait_s": 1.7583002772880718e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 514400, "loss": 0.5038048624992371, "lr": 0.0003, "grad_norm": 0.26869483123555776, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75290707061, "tokens_per_second": 150772.69595658488, "data_wait_s": 1.7244998161913827e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 514500, "loss": 0.5570145845413208, "lr": 0.0003, "grad_norm": 0.1675595931676757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75305303347, "tokens_per_second": 152018.73569825958, "data_wait_s": 1.6735997633077204e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e1a3c4909f972d1dfa2626fabffa9d2722a7316fa86396b544ce8d06e9eaaa7
-size 626772
+oid sha256:881df4fd5a0b843fc4ba9a349e70ce59f8762c17a08d968d7463f5b7c8e44397
+size 628890