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": 75046.4854780052,
-      "required_pages_per_second_upper_estimate": 0.0065048391576549725
+      "required_token_positions_per_second": 74921.12956754336,
+      "required_pages_per_second_upper_estimate": 0.0064939736250477905
     },
     {
       "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": 9380.81068475065,
-      "required_pages_per_second_upper_estimate": 0.0037421456377655375
+      "required_token_positions_per_second": 9365.14119594292,
+      "required_pages_per_second_upper_estimate": 0.003735894844400399
     },
     {
       "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": 18761.6213695013,
-      "required_pages_per_second_upper_estimate": 0.004340170910575522
+      "required_token_positions_per_second": 18730.28239188584,
+      "required_pages_per_second_upper_estimate": 0.0043329211896513545
     },
     {
       "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": 28142.43205425195,
-      "required_pages_per_second_upper_estimate": 0.0017137264601740974
+      "required_token_positions_per_second": 28095.42358782876,
+      "required_pages_per_second_upper_estimate": 0.0017108638912032878
     },
     {
       "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": 28142.43205425195,
-      "required_pages_per_second_upper_estimate": 0.0021664823227742704
+      "required_token_positions_per_second": 28095.42358782876,
+      "required_pages_per_second_upper_estimate": 0.00216286348090125
     },
     {
       "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": 28142.43205425195,
-      "required_pages_per_second_upper_estimate": 0.349434819452574
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28095.42358782876,
+      "required_pages_per_second_upper_estimate": 0.3488511316268145
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.2684761333126744,
-    "producer_p95_seconds": 0.3894732529952307,
-    "gpu_update_seconds": 0.9824310829533567,
-    "producer_batches_per_second": 3.724725872878144,
-    "gpu_batches_per_second": 1.0178831038140896,
-    "producer_headroom": 3.6592864729960617,
+    "producer_batch_seconds": 0.7118227192659106,
+    "producer_p95_seconds": 0.3801378630014369,
+    "gpu_update_seconds": 0.9840748587957724,
+    "producer_batches_per_second": 1.404844173885432,
+    "gpu_batches_per_second": 1.0161828554625565,
+    "producer_headroom": 1.38247183204637,
     "sustainable": true,
-    "buffer_seconds": 60.91072714310811,
+    "buffer_seconds": 60.02856638654212,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342650.8197081001,
-    "upload_bytes_per_second": 3962746.0647738087,
+    "generation_bytes_per_second": 302091.04154830106,
+    "upload_bytes_per_second": 4074442.7903771154,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.5248894065118,
+    "estimated_full_shards_per_hour": 64.82170519673132,
     "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": 0.8848312870031805,
-      "bytes": 821123,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.1538495610002428,
       "bytes": 7799438,
@@ -562,6 +556,12 @@
       "bytes": 1059989,
       "pages": 33,
       "shards": 1
+    },
+    {
+      "seconds": 1.474424235006154,
+      "bytes": 14927789,
+      "pages": 21,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -339,3 +339,4 @@
 {"step": 348900, "loss": 0.5869564954191446, "lr": 0.0003, "grad_norm": 0.15196084157394468, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51069817667, "tokens_per_second": 149364.34447103136, "data_wait_s": 1.632799830986187e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349000, "loss": 0.6107743624597788, "lr": 0.0003, "grad_norm": 0.1425909478465102, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51084555829, "tokens_per_second": 149890.57271048252, "data_wait_s": 1.2324999261181802e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 349100, "loss": 0.584244143217802, "lr": 0.0003, "grad_norm": 0.15031535589287243, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51099287026, "tokens_per_second": 131858.31811939453, "data_wait_s": 1.5056997654028237e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 349200, "loss": 0.5608677826821804, "lr": 0.0003, "grad_norm": 0.1469427349181762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 51113978707, "tokens_per_second": 149871.6101259567, "data_wait_s": 1.3037999451626092e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe6f8f5ae5a12862c5396e88cc5d3e409def7fbde89145d9350e50fd1f2c5a84
-size 727101
+oid sha256:a47ea14df1a61477d4df4d1f74c8da9e2f6109d5a827bd482423ef13780f91e5
+size 729219