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": 75352.79662403779,
-      "required_pages_per_second_upper_estimate": 0.006531389431454575
+      "required_token_positions_per_second": 75643.47127921894,
+      "required_pages_per_second_upper_estimate": 0.006556584373857499
     },
     {
       "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": 9419.099578004723,
-      "required_pages_per_second_upper_estimate": 0.003757419649754557
+      "required_token_positions_per_second": 9455.433909902367,
+      "required_pages_per_second_upper_estimate": 0.0037719139579952
     },
     {
       "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": 18838.199156009447,
-      "required_pages_per_second_upper_estimate": 0.00435788583376117
+      "required_token_positions_per_second": 18910.867819804735,
+      "required_pages_per_second_upper_estimate": 0.004374696450205408
     },
     {
       "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": 28257.298734014166,
-      "required_pages_per_second_upper_estimate": 0.0017207212383127236
+      "required_token_positions_per_second": 28366.301729707102,
+      "required_pages_per_second_upper_estimate": 0.0017273589488559034
     },
     {
       "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": 28257.298734014166,
-      "required_pages_per_second_upper_estimate": 0.0021753250777536864
+      "required_token_positions_per_second": 28366.301729707102,
+      "required_pages_per_second_upper_estimate": 0.002183716430101733
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28257.298734014166,
-      "required_pages_per_second_upper_estimate": 0.35086107918117343
+      "required_token_positions_per_second": 28366.301729707102,
+      "required_pages_per_second_upper_estimate": 0.35221453157812066
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2481843025313708,
-    "producer_p95_seconds": 0.29789283699938096,
-    "gpu_update_seconds": 0.9784374741637729,
-    "producer_batches_per_second": 4.029263695569943,
-    "gpu_batches_per_second": 1.0220377146272486,
-    "producer_headroom": 3.9423825930332446,
+    "producer_batch_seconds": 0.06656445685169388,
+    "producer_p95_seconds": 0.22578877799969632,
+    "gpu_update_seconds": 0.9746776391032023,
+    "producer_batches_per_second": 15.023032520613933,
+    "gpu_batches_per_second": 1.0259802419598922,
+    "producer_headroom": 14.642613869362618,
     "sustainable": true,
-    "buffer_seconds": 60.66312339815392,
+    "buffer_seconds": 60.43001362439854,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 421816.9297625074,
-    "upload_bytes_per_second": 5970901.868635902,
+    "generation_bytes_per_second": 420752.1972591669,
+    "upload_bytes_per_second": 5971096.05078665,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.51209373146455,
+    "estimated_full_shards_per_hour": 90.28362692195182,
     "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": 0.9569115700014663,
-      "bytes": 1242347,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.07777969900053,
       "bytes": 7140650,
@@ -562,6 +556,12 @@
       "bytes": 11732086,
       "pages": 34,
       "shards": 3
+    },
+    {
+      "seconds": 0.8983965400002489,
+      "bytes": 906815,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -140,3 +140,4 @@
 {"step": 99000, "loss": 0.5780111197382212, "lr": 0.0003, "grad_norm": 0.13925460440210524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14486645414, "tokens_per_second": 148265.99686362976, "data_wait_s": 1.2526001228252426e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99100, "loss": 0.6534293703734875, "lr": 0.0003, "grad_norm": 0.15429198608295136, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14501253596, "tokens_per_second": 148392.02871361803, "data_wait_s": 1.5487999917240813e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 99200, "loss": 0.6560120675712824, "lr": 0.0003, "grad_norm": 0.1448230448036412, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14515937038, "tokens_per_second": 149260.46656382762, "data_wait_s": 1.048400008585304e-05, "gpu_peak_gb": 85.46991729736328}
+{"step": 99300, "loss": 0.5981705710291862, "lr": 0.0003, "grad_norm": 0.14313228307489784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14530706160, "tokens_per_second": 149255.5627441101, "data_wait_s": 1.6499998309882358e-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 sha256:84be40ed1eda144bf9d26b8927942b76b11234e406551e428c6b7695ce40320f
-size 302487
+oid sha256:75c2c8aa26756b41534a7fba3df460251b2a40d26a0d44b6fb9588816bc6d5de
+size 304605