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": 77095.47018256475,
-      "required_pages_per_second_upper_estimate": 0.005333643513366353
+      "required_token_positions_per_second": 76890.77989178842,
+      "required_pages_per_second_upper_estimate": 0.005319482564103537
     },
     {
       "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": 9636.933772820594,
-      "required_pages_per_second_upper_estimate": 0.004034782928618808
+      "required_token_positions_per_second": 9611.347486473553,
+      "required_pages_per_second_upper_estimate": 0.004024070484827719
     },
     {
       "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": 19273.867545641187,
-      "required_pages_per_second_upper_estimate": 0.004014238323689123
+      "required_token_positions_per_second": 19222.694972947105,
+      "required_pages_per_second_upper_estimate": 0.004003580426308439
     },
     {
       "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": 28910.80131846178,
-      "required_pages_per_second_upper_estimate": 0.0017331583948042048
+      "required_token_positions_per_second": 28834.042459420656,
+      "required_pages_per_second_upper_estimate": 0.0017285568184086807
     },
     {
       "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": 28910.80131846178,
-      "required_pages_per_second_upper_estimate": 0.002247687637334634
+      "required_token_positions_per_second": 28834.042459420656,
+      "required_pages_per_second_upper_estimate": 0.0022417199736707257
     },
     {
       "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": 28910.80131846178,
-      "required_pages_per_second_upper_estimate": 0.3708367173132949
+      "required_token_positions_per_second": 28834.042459420656,
+      "required_pages_per_second_upper_estimate": 0.3698521370867569
     }
   ],
   "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": 0.061708173382768905,
-    "producer_p95_seconds": 0.20968927200010512,
-    "gpu_update_seconds": 0.9563207776722749,
-    "producer_batches_per_second": 16.20530871651494,
-    "gpu_batches_per_second": 1.0456742375022345,
-    "producer_headroom": 15.49747343419686,
+    "producer_batch_seconds": 0.1304158580234258,
+    "producer_p95_seconds": 0.23044491799919342,
+    "gpu_update_seconds": 0.9588665910758152,
+    "producer_batches_per_second": 7.667779173146076,
+    "gpu_batches_per_second": 1.0428979477510365,
+    "producer_headroom": 7.352377276876711,
     "sustainable": true,
-    "buffer_seconds": 60.248208993353316,
+    "buffer_seconds": 60.40859523777635,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 158,
+    "upload_pages_per_commit": 162,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 610824.6017554874,
-    "upload_bytes_per_second": 5097172.906427652,
+    "generation_bytes_per_second": 595690.6470669004,
+    "upload_bytes_per_second": 5165362.117473496,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.53437013386949,
+    "estimated_full_shards_per_hour": 63.910672946001334,
     "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": 1.0023074259997884,
-      "bytes": 1140621,
-      "pages": 35,
-      "shards": 1
-    },
     {
       "seconds": 1.374354456999754,
       "bytes": 11971022,
@@ -562,6 +556,12 @@
       "bytes": 6307877,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1178816689989617,
+      "bytes": 7521323,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -139,3 +139,4 @@
 {"step": 454100, "loss": 0.5607672929763794, "lr": 0.0003, "grad_norm": 0.14591290389108755, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66466172541, "tokens_per_second": 151110.56218026747, "data_wait_s": 1.583700031915214e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 454200, "loss": 0.5829521417617798, "lr": 0.0003, "grad_norm": 0.14400335327144934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66480830643, "tokens_per_second": 152923.4861655191, "data_wait_s": 1.7417998606106266e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 454300, "loss": 0.559476375579834, "lr": 0.0003, "grad_norm": 0.15328640335116137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66495484954, "tokens_per_second": 152184.67743569327, "data_wait_s": 1.4470999303739518e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 454400, "loss": 0.6459651589393616, "lr": 0.0003, "grad_norm": 0.17926437166238893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66510301729, "tokens_per_second": 154395.115004786, "data_wait_s": 1.0058998668682761e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bbfdd7423c8fdd71f5111cce205609040c2c2f5a2c8f5072302a68282613ef91
-size 300369
+oid sha256:5bad8f0541812186d2e9793d8f3e74490d75ee9a55e536d91b06119ca808f010
+size 302487