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": 76923.73572403575,
-      "required_pages_per_second_upper_estimate": 0.005321762525046473
+      "required_token_positions_per_second": 77168.29278915029,
+      "required_pages_per_second_upper_estimate": 0.0053386815502616675
     },
     {
       "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": 9615.466965504469,
-      "required_pages_per_second_upper_estimate": 0.004025795224673459
+      "required_token_positions_per_second": 9646.036598643786,
+      "required_pages_per_second_upper_estimate": 0.0040385940917023605
     },
     {
       "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": 19230.933931008938,
-      "required_pages_per_second_upper_estimate": 0.004005296383996783
+      "required_token_positions_per_second": 19292.073197287573,
+      "required_pages_per_second_upper_estimate": 0.004018030080811745
     },
     {
       "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": 28846.40089651341,
-      "required_pages_per_second_upper_estimate": 0.0017292976878421505
+      "required_token_positions_per_second": 28938.10979593136,
+      "required_pages_per_second_upper_estimate": 0.0017347954963308752
     },
     {
       "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": 28846.40089651341,
-      "required_pages_per_second_upper_estimate": 0.002242680787795667
+      "required_token_positions_per_second": 28938.10979593136,
+      "required_pages_per_second_upper_estimate": 0.0022498107513405953
     },
     {
       "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": 28846.40089651341,
-      "required_pages_per_second_upper_estimate": 0.3700106578483268
+      "required_token_positions_per_second": 28938.10979593136,
+      "required_pages_per_second_upper_estimate": 0.371187001140716
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.0790830256640902,
-    "producer_p95_seconds": 0.2122634030001791,
-    "gpu_update_seconds": 0.9584557913892733,
-    "producer_batches_per_second": 12.644938551637601,
-    "gpu_batches_per_second": 1.0433449398333843,
-    "producer_headroom": 12.119614586578548,
+    "producer_batch_seconds": 0.23938429012494566,
+    "producer_p95_seconds": 0.2290029340001638,
+    "gpu_update_seconds": 0.9554183115265964,
+    "producer_batches_per_second": 4.17738356797789,
+    "gpu_batches_per_second": 1.0466619573181193,
+    "producer_headroom": 3.991148755116385,
     "sustainable": true,
-    "buffer_seconds": 60.38271485752421,
+    "buffer_seconds": 60.19135362617557,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 247642.8912956146,
-    "upload_bytes_per_second": 5140511.247586042,
+    "generation_bytes_per_second": 372119.1062006688,
+    "upload_bytes_per_second": 5196451.587772348,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.13839964057282,
+    "estimated_full_shards_per_hour": 79.8480977012162,
     "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": 1.873195456000758,
-      "bytes": 6027495,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.351263048999499,
       "bytes": 5825113,
@@ -562,6 +556,12 @@
       "bytes": 2291279,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.055462077001721,
+      "bytes": 6384191,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -123,3 +123,4 @@
 {"step": 657300, "loss": 0.5231503844261169, "lr": 0.0001, "grad_norm": 0.1778715153824788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96211010820, "tokens_per_second": 153098.66837463455, "data_wait_s": 9.149000106845051e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 657400, "loss": 0.5541929006576538, "lr": 0.0001, "grad_norm": 0.22244126197697556, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96225596595, "tokens_per_second": 152390.4617840014, "data_wait_s": 9.798999599297531e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 657500, "loss": 0.5422726273536682, "lr": 0.0001, "grad_norm": 0.184700603723759, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96240272694, "tokens_per_second": 152578.73015433078, "data_wait_s": 9.624998710933141e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 657600, "loss": 0.5721409320831299, "lr": 0.0001, "grad_norm": 0.23453476625534417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96254934317, "tokens_per_second": 153055.72192716855, "data_wait_s": 9.101000614464283e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae7b7e3b358561ad7768b033f32ca4d8bba7292c8d810a657dd4364152e16ab6
-size 266481
+oid sha256:80ebc494c13faa1e1e27461518adc6d788821baa0b5840938e4dedfa858e8949
+size 268599