Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75280.14148829899,
-      "required_pages_per_second_upper_estimate": 0.0065250918684314945
+      "required_token_positions_per_second": 74464.66887455113,
+      "required_pages_per_second_upper_estimate": 0.006454408769068274
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9410.017686037374,
-      "required_pages_per_second_upper_estimate": 0.003753796747262396
+      "required_token_positions_per_second": 9308.083609318892,
+      "required_pages_per_second_upper_estimate": 0.0037131337200091317
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18820.035372074748,
-      "required_pages_per_second_upper_estimate": 0.004353683962019557
+      "required_token_positions_per_second": 18616.167218637784,
+      "required_pages_per_second_upper_estimate": 0.004306522652678872
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28230.05305811212,
-      "required_pages_per_second_upper_estimate": 0.001719062119597303
+      "required_token_positions_per_second": 27924.250827956672,
+      "required_pages_per_second_upper_estimate": 0.0017004403682011418
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28230.05305811212,
-      "required_pages_per_second_upper_estimate": 0.0021732276302018887
+      "required_token_positions_per_second": 27924.250827956672,
+      "required_pages_per_second_upper_estimate": 0.00214968612800977
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28230.05305811212,
-      "required_pages_per_second_upper_estimate": 0.3505227790718815
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27924.250827956672,
+      "required_pages_per_second_upper_estimate": 0.3467257388275783
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.551446997500193,
-    "producer_p95_seconds": 0.3606329719987116,
-    "gpu_update_seconds": 0.9793817936893724,
-    "producer_batches_per_second": 1.813410907182698,
-    "gpu_batches_per_second": 1.021052266280097,
-    "producer_headroom": 1.776021626972463,
+    "producer_batch_seconds": 0.41645254064832216,
+    "producer_p95_seconds": 0.3593799950031098,
+    "gpu_update_seconds": 0.9901071355626092,
+    "producer_batches_per_second": 2.401234000021291,
+    "gpu_batches_per_second": 1.009991711080609,
+    "producer_headroom": 2.377478917576627,
     "sustainable": true,
-    "buffer_seconds": 60.721671208741085,
+    "buffer_seconds": 60.39653526931916,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 75,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 272253.27319435315,
-    "upload_bytes_per_second": 5958677.705292,
+    "generation_bytes_per_second": 274396.3375249854,
+    "upload_bytes_per_second": 5984801.876030966,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 58.41921469567247,
+    "estimated_full_shards_per_hour": 58.879066413041805,
     "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": 1.300397458999214,
-      "bytes": 11506344,
-      "pages": 35,
-      "shards": 4
-    },
     {
       "seconds": 1.0356865150024532,
       "bytes": 6075433,
@@ -562,6 +556,12 @@
       "bytes": 6000035,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.173013809006079,
+      "bytes": 12591906,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790485060735632541-7579de36/train.jsonl

@@ -274,3 +274,4 @@
 {"step": 112400, "loss": 0.6337862759828568, "lr": 0.0003, "grad_norm": 0.17847442054156323, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16447657916, "tokens_per_second": 150064.66074569745, "data_wait_s": 1.2643999070860445e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 112500, "loss": 0.5730115100741386, "lr": 0.0003, "grad_norm": 0.12394509323280861, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16462338797, "tokens_per_second": 133919.43582881003, "data_wait_s": 1.099500514101237e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 112600, "loss": 0.6193175911903381, "lr": 0.0003, "grad_norm": 0.14174843665291517, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16477043167, "tokens_per_second": 148836.8831800993, "data_wait_s": 1.8449994968250394e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 112700, "loss": 0.6256785094738007, "lr": 0.0003, "grad_norm": 0.14013686517648433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 16491584949, "tokens_per_second": 147834.45150944236, "data_wait_s": 1.581400283612311e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790485060.0b09bbb2-f535-42d2-bd00-cbae7fb94f51-r2qrq.28895.0.1790485060735632541-7579de36

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:40bbfbdeca1819d9e935e99f94e0f3b08b70b8c60e402be7b1238ef837ba9826
-size 588648
+oid sha256:d2c340c22eef9773b840e4ab7d2fb15dcb8301b2927b283d4c7b93f05f064ae9
+size 590766