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": 76790.5686850659,
-      "required_pages_per_second_upper_estimate": 0.005312549720300438
+      "required_token_positions_per_second": 77024.4758884758,
+      "required_pages_per_second_upper_estimate": 0.005328731963371822
     },
     {
       "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": 9598.821085633237,
-      "required_pages_per_second_upper_estimate": 0.004018825942376874
+      "required_token_positions_per_second": 9628.059486059476,
+      "required_pages_per_second_upper_estimate": 0.004031067450068109
     },
     {
       "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": 19197.642171266474,
-      "required_pages_per_second_upper_estimate": 0.003998362588405173
+      "required_token_positions_per_second": 19256.11897211895,
+      "required_pages_per_second_upper_estimate": 0.004010541763885801
     },
     {
       "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": 28796.463256899708,
-      "required_pages_per_second_upper_estimate": 0.0017263040025976698
+      "required_token_positions_per_second": 28884.178458178427,
+      "required_pages_per_second_upper_estimate": 0.0017315623949809716
     },
     {
       "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": 28796.463256899708,
-      "required_pages_per_second_upper_estimate": 0.002238798355968165
+      "required_token_positions_per_second": 28884.178458178427,
+      "required_pages_per_second_upper_estimate": 0.0022456178270491896
     },
     {
       "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": 28796.463256899708,
-      "required_pages_per_second_upper_estimate": 0.36937011142622217
+      "required_token_positions_per_second": 28884.178458178427,
+      "required_pages_per_second_upper_estimate": 0.37049522784698025
     }
   ],
   "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": 0.10840794935928955,
-    "producer_p95_seconds": 0.21280910899986338,
-    "gpu_update_seconds": 0.9601179059159449,
-    "producer_batches_per_second": 9.224415791555689,
-    "gpu_batches_per_second": 1.0415387462709675,
-    "producer_headroom": 8.856526773086422,
+    "producer_batch_seconds": 0.5100505060154887,
+    "producer_p95_seconds": 0.21336384899950644,
+    "gpu_update_seconds": 0.9572022288960617,
+    "producer_batches_per_second": 1.9605901537320167,
+    "gpu_batches_per_second": 1.0447113157616619,
+    "producer_headroom": 1.8766812651039586,
     "sustainable": true,
-    "buffer_seconds": 60.48742807270453,
+    "buffer_seconds": 60.30374042045189,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 331807.6524095087,
-    "upload_bytes_per_second": 5331855.523753104,
+    "generation_bytes_per_second": 301917.61175707204,
+    "upload_bytes_per_second": 5349415.495151054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.19819812025018,
+    "estimated_full_shards_per_hour": 64.78449120077248,
     "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": 0.9793791169995529,
-      "bytes": 5849196,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 0.8912132279992875,
       "bytes": 2690813,
@@ -562,6 +556,12 @@
       "bytes": 861206,
       "pages": 26,
       "shards": 1
+    },
+    {
+      "seconds": 1.055939247000424,
+      "bytes": 7659150,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -131,3 +131,4 @@
 {"step": 570200, "loss": 0.587658166885376, "lr": 0.0003, "grad_norm": 0.1683831684998894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83460408739, "tokens_per_second": 151471.25496444936, "data_wait_s": 1.2197000614833087e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 570300, "loss": 0.6701946258544922, "lr": 0.0003, "grad_norm": 0.2391639645788086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83474990127, "tokens_per_second": 151590.5282242714, "data_wait_s": 1.3880000551580451e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 570400, "loss": 0.5762056708335876, "lr": 0.0003, "grad_norm": 0.14501432976856904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83489649800, "tokens_per_second": 153394.0111943038, "data_wait_s": 8.924000212573446e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 570500, "loss": 0.5844987034797668, "lr": 0.0003, "grad_norm": 0.1595094048445295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83504268929, "tokens_per_second": 137069.08425664963, "data_wait_s": 1.3202999980421737e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5de0134794ae9de5e9083485c7960ae9ffaf0b55e22e3914dcf6c171ee4fd390
-size 284208
+oid sha256:eb1da03027cd91663f4043aafa417e15a460c8f3ccaf12bc0ac74c562dac11df
+size 286326