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": 75645.33855517661,
-      "required_pages_per_second_upper_estimate": 0.006556746224604943
+      "required_token_positions_per_second": 74881.73535716788,
+      "required_pages_per_second_upper_estimate": 0.0064905590347361574
     },
     {
       "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": 9455.667319397076,
-      "required_pages_per_second_upper_estimate": 0.0037720070685324223
+      "required_token_positions_per_second": 9360.216919645985,
+      "required_pages_per_second_upper_estimate": 0.00373393047696106
     },
     {
       "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": 18911.334638794153,
-      "required_pages_per_second_upper_estimate": 0.0043748044405628765
+      "required_token_positions_per_second": 18720.43383929197,
+      "required_pages_per_second_upper_estimate": 0.004330642900337368
     },
     {
       "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": 28367.00195819123,
-      "required_pages_per_second_upper_estimate": 0.001727401589096772
+      "required_token_positions_per_second": 28080.650758937958,
+      "required_pages_per_second_upper_estimate": 0.001709964303430881
     },
     {
       "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": 28367.00195819123,
-      "required_pages_per_second_upper_estimate": 0.0021837703356287976
+      "required_token_positions_per_second": 28080.650758937958,
+      "required_pages_per_second_upper_estimate": 0.002161726227639428
     },
     {
       "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": 28367.00195819123,
-      "required_pages_per_second_upper_estimate": 0.35222322607237955
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28080.650758937958,
+      "required_pages_per_second_upper_estimate": 0.34866770253346857
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.08588451295312183,
-    "producer_p95_seconds": 0.132505854999863,
-    "gpu_update_seconds": 0.9746535795622345,
-    "producer_batches_per_second": 11.64354277174312,
-    "gpu_batches_per_second": 1.0260055685109675,
-    "producer_headroom": 11.348420641265413,
+    "producer_batch_seconds": 0.1693122093046977,
+    "producer_p95_seconds": 0.13976951599988752,
+    "gpu_update_seconds": 0.984592566509512,
+    "producer_batches_per_second": 5.906248604909405,
+    "gpu_batches_per_second": 1.015648537287976,
+    "producer_headroom": 5.815248472350976,
     "sustainable": true,
-    "buffer_seconds": 60.428521932858544,
+    "buffer_seconds": 60.06014655708023,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 142,
+    "upload_pages_per_commit": 144,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 768810.9648677881,
-    "upload_bytes_per_second": 4708605.836080916,
+    "generation_bytes_per_second": 693026.4252446456,
+    "upload_bytes_per_second": 5155872.122164343,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.4844680286657,
+    "estimated_full_shards_per_hour": 74.35366901399863,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 2622281,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.0029198830000041,
+      "bytes": 11222438,
+      "pages": 33,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -8,3 +8,4 @@
 {"step": 800, "loss": 1.3345604501664639, "lr": 0.00023999999999999998, "grad_norm": 0.8847908940671446, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 117723562, "tokens_per_second": 148830.22167264167, "data_wait_s": 1.11660001493874e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 900, "loss": 1.2485260888934135, "lr": 0.00026999999999999995, "grad_norm": 0.5126180461192602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 132386398, "tokens_per_second": 148764.82655347895, "data_wait_s": 2.0052000081705046e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 1000, "loss": 1.1477682627737522, "lr": 0.0003, "grad_norm": 0.40807168767980584, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 146947989, "tokens_per_second": 149593.43493141377, "data_wait_s": 1.4937000059944694e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 1100, "loss": 1.0527566373348236, "lr": 0.0003, "grad_norm": 0.35079080124550893, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 161575502, "tokens_per_second": 148849.57722070834, "data_wait_s": 1.6026000139390817e-05, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:91aab8257a64f0356dcef4bb0ce36d08765a7829399bd6a3317fbf7d902a8be1
-size 20956
+oid sha256:7f65c8f69536fa008e364d831a8c3d8498fc4dc663b54eac8d52fa232715f602
+size 23039