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": 76366.33618496417,
-      "required_pages_per_second_upper_estimate": 0.005283200331588382
+      "required_token_positions_per_second": 76036.75667570926,
+      "required_pages_per_second_upper_estimate": 0.005260399256408307
     },
     {
       "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": 9545.79202312052,
-      "required_pages_per_second_upper_estimate": 0.003996623781275548
+      "required_token_positions_per_second": 9504.594584463657,
+      "required_pages_per_second_upper_estimate": 0.003979375274010266
     },
     {
       "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": 19091.58404624104,
-      "required_pages_per_second_upper_estimate": 0.003976273477903218
+      "required_token_positions_per_second": 19009.189168927314,
+      "required_pages_per_second_upper_estimate": 0.003959112797857805
     },
     {
       "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": 28637.376069361562,
-      "required_pages_per_second_upper_estimate": 0.0017167669686168282
+      "required_token_positions_per_second": 28513.78375339097,
+      "required_pages_per_second_upper_estimate": 0.0017093577979889307
     },
     {
       "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": 28637.376069361562,
-      "required_pages_per_second_upper_estimate": 0.0022264300268876602
+      "required_token_positions_per_second": 28513.78375339097,
+      "required_pages_per_second_upper_estimate": 0.0022168212679460927
     },
     {
       "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": 28637.376069361562,
-      "required_pages_per_second_upper_estimate": 0.3673295117989965
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28513.78375339097,
+      "required_pages_per_second_upper_estimate": 0.3657442022728155
     }
   ],
   "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": 0.4998368583280808,
-    "producer_p95_seconds": 0.2842813349998323,
-    "gpu_update_seconds": 0.9654515809351657,
-    "producer_batches_per_second": 2.000652779678813,
-    "gpu_batches_per_second": 1.0357847247309593,
-    "producer_headroom": 1.931533389043244,
+    "producer_batch_seconds": 0.1304749311640876,
+    "producer_p95_seconds": 0.2815488330015796,
+    "gpu_update_seconds": 0.9696363077983992,
+    "producer_batches_per_second": 7.664307549949057,
+    "gpu_batches_per_second": 1.0313145165433655,
+    "producer_headroom": 7.431590874563998,
     "sustainable": true,
-    "buffer_seconds": 60.82344959891544,
+    "buffer_seconds": 60.11745108350075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307269.6231738124,
-    "upload_bytes_per_second": 4736748.722279621,
+    "generation_bytes_per_second": 320750.0612110643,
+    "upload_bytes_per_second": 4793516.517513572,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.93290826235561,
+    "estimated_full_shards_per_hour": 68.82549645661302,
     "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": 1.0866290939993632,
-      "bytes": 1216935,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.176057447999483,
       "bytes": 7472528,
@@ -562,6 +556,12 @@
       "bytes": 5933529,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.2705231469990395,
+      "bytes": 7122874,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -307,3 +307,4 @@
 {"step": 790700, "loss": 0.5069711804389954, "lr": 3e-05, "grad_norm": 0.2505566734304559, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115735872939, "tokens_per_second": 152687.89437593988, "data_wait_s": 9.497998689766973e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 790800, "loss": 0.534726083278656, "lr": 3e-05, "grad_norm": 0.2632059355800747, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115750515759, "tokens_per_second": 151996.63088258673, "data_wait_s": 1.0162999387830496e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 790900, "loss": 0.5075185894966125, "lr": 3e-05, "grad_norm": 0.25775752807402286, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115765132292, "tokens_per_second": 151411.71402798707, "data_wait_s": 8.085000445134938e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 791000, "loss": 0.507858395576477, "lr": 3e-05, "grad_norm": 0.2734002704141276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 115779772392, "tokens_per_second": 152049.74784515027, "data_wait_s": 1.0251002095174044e-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:d7b0f6feb0ae9e9faf61a7bccdb26f1e6883873dad33651de8f032f84a95d556
-size 659325
+oid sha256:8efda08f701f1ef7b2a73c08361193fa9f76e5b67f38ade4d0698121ad365fd0
+size 661443