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": 75992.76807919395,
-      "required_pages_per_second_upper_estimate": 0.006586860535194687
+      "required_token_positions_per_second": 73292.50666727634,
+      "required_pages_per_second_upper_estimate": 0.006352808719759648
     },
     {
       "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": 9499.096009899244,
-      "required_pages_per_second_upper_estimate": 0.0037893314224905234
+      "required_token_positions_per_second": 9161.563333409542,
+      "required_pages_per_second_upper_estimate": 0.0036546845912755472
     },
     {
       "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": 18998.192019798487,
-      "required_pages_per_second_upper_estimate": 0.004394897367020533
+      "required_token_positions_per_second": 18323.126666819084,
+      "required_pages_per_second_upper_estimate": 0.004238732878353182
     },
     {
       "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": 28497.288029697735,
-      "required_pages_per_second_upper_estimate": 0.001735335327293329
+      "required_token_positions_per_second": 27484.690000228624,
+      "required_pages_per_second_upper_estimate": 0.0016736734199899344
     },
     {
       "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": 28497.288029697735,
-      "required_pages_per_second_upper_estimate": 0.002193800118068297
+      "required_token_positions_per_second": 27484.690000228624,
+      "required_pages_per_second_upper_estimate": 0.0021158475187090164
     },
     {
       "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": 28497.288029697735,
-      "required_pages_per_second_upper_estimate": 0.35384094304105856
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27484.690000228624,
+      "required_pages_per_second_upper_estimate": 0.3412678644626522
     }
   ],
   "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": 60,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10566539140666009,
-    "producer_p95_seconds": 0.4347988439985784,
-    "gpu_update_seconds": 0.9701975841065063,
-    "producer_batches_per_second": 9.463836613744563,
-    "gpu_batches_per_second": 1.0307178830185812,
-    "producer_headroom": 9.181791419033676,
+    "producer_batch_seconds": 0.25148598814905654,
+    "producer_p95_seconds": 0.3997792459995253,
+    "gpu_update_seconds": 1.005941853438042,
+    "producer_batches_per_second": 3.976364676855463,
+    "gpu_batches_per_second": 0.9940932436425284,
+    "producer_headroom": 3.9999916529815454,
     "sustainable": true,
-    "buffer_seconds": 60.152250214603384,
+    "buffer_seconds": 60.35651120628252,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343158.7487764553,
-    "upload_bytes_per_second": 5941447.287806251,
+    "generation_bytes_per_second": 341950.87908101245,
+    "upload_bytes_per_second": 5760194.18075138,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.63387916059727,
+    "estimated_full_shards_per_hour": 73.37469844172269,
     "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": 1.2370568289989023,
-      "bytes": 16078869,
-      "pages": 34,
-      "shards": 4
-    },
     {
       "seconds": 0.8470632479984488,
       "bytes": 947755,
@@ -562,6 +556,12 @@
       "bytes": 11872287,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 1.1487145990031422,
+      "bytes": 2742520,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 154600, "loss": 0.6058161295950413, "lr": 0.0003, "grad_norm": 0.14773088963732545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22625939930, "tokens_per_second": 149532.91651421343, "data_wait_s": 1.1928001185879111e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 154700, "loss": 0.6039471216499805, "lr": 0.0003, "grad_norm": 0.14264851109627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22640513205, "tokens_per_second": 149201.79804365436, "data_wait_s": 1.2743999832309783e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 154800, "loss": 0.6642542071640491, "lr": 0.0003, "grad_norm": 0.1520688864962238, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22655207482, "tokens_per_second": 149167.29114741713, "data_wait_s": 8.737995813135058e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 154900, "loss": 0.610124297440052, "lr": 0.0003, "grad_norm": 0.1712330672449007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 22669651712, "tokens_per_second": 146885.69582243974, "data_wait_s": 1.0073003068100661e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b17f59cc59f59deb685027bd90d9b03f7f7d1ec4e8093f361138e39b37f15eb
-size 848610
+oid sha256:39e2cf8259344cec8de2785f5747cd373769d94fbe428f4a6816d5ca3950d1da
+size 850728