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": 74744.03122129888,
-      "required_pages_per_second_upper_estimate": 0.006478623189246974
+      "required_token_positions_per_second": 75352.79662403779,
+      "required_pages_per_second_upper_estimate": 0.006531389431454575
     },
     {
       "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": 9343.00390266236,
-      "required_pages_per_second_upper_estimate": 0.0037270639471287535
+      "required_token_positions_per_second": 9419.099578004723,
+      "required_pages_per_second_upper_estimate": 0.003757419649754557
     },
     {
       "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": 18686.00780532472,
-      "required_pages_per_second_upper_estimate": 0.00432267904325655
+      "required_token_positions_per_second": 18838.199156009447,
+      "required_pages_per_second_upper_estimate": 0.00435788583376117
     },
     {
       "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": 28029.01170798708,
-      "required_pages_per_second_upper_estimate": 0.0017068197561571343
+      "required_token_positions_per_second": 28257.298734014166,
+      "required_pages_per_second_upper_estimate": 0.0017207212383127236
     },
     {
       "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": 28029.01170798708,
-      "required_pages_per_second_upper_estimate": 0.0021577509105511873
+      "required_token_positions_per_second": 28257.298734014166,
+      "required_pages_per_second_upper_estimate": 0.0021753250777536864
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 28029.01170798708,
-      "required_pages_per_second_upper_estimate": 0.3480265183454447
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28257.298734014166,
+      "required_pages_per_second_upper_estimate": 0.35086107918117343
     }
   ],
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.09684319732809854,
-    "producer_p95_seconds": 0.29584928700205637,
-    "gpu_update_seconds": 0.9864065236421266,
-    "producer_batches_per_second": 10.325970513055905,
-    "gpu_batches_per_second": 1.0137808054104123,
-    "producer_headroom": 10.185604677014583,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.17079794216972,
+    "buffer_seconds": 60.66312339815392,
     "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": 403596.85687995487,
-    "upload_bytes_per_second": 5936136.164543037,
+    "generation_bytes_per_second": 421816.9297625074,
+    "upload_bytes_per_second": 5970901.868635902,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.60249023245797,
+    "estimated_full_shards_per_hour": 90.51209373146455,
     "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": 1.0362804550004512,
-      "bytes": 7507968,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9569115700014663,
       "bytes": 1242347,
@@ -562,6 +556,12 @@
       "bytes": 2562080,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.3296433830000751,
+      "bytes": 11732086,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -139,3 +139,4 @@
 {"step": 98900, "loss": 0.6071541821584105, "lr": 0.0003, "grad_norm": 0.13787328096811033, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 14472080736, "tokens_per_second": 146747.2228629315, "data_wait_s": 1.5404002624563873e-05, "gpu_peak_gb": 85.46991729736328}
 {"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}

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:ebc96a16c725367198da260747ae903d5e0b6e88957ba1d7154cb01bb9095e09
-size 300369
+oid sha256:84be40ed1eda144bf9d26b8927942b76b11234e406551e428c6b7695ce40320f
+size 302487