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": 75881.21861129551,
-      "required_pages_per_second_upper_estimate": 0.006577191710036792
+      "required_token_positions_per_second": 75862.55319913692,
+      "required_pages_per_second_upper_estimate": 0.00657557383941794
     },
     {
       "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": 9485.152326411939,
-      "required_pages_per_second_upper_estimate": 0.003783769078670791
+      "required_token_positions_per_second": 9482.819149892115,
+      "required_pages_per_second_upper_estimate": 0.003782838339672936
     },
     {
       "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": 18970.304652823877,
-      "required_pages_per_second_upper_estimate": 0.004388446115471852
+      "required_token_positions_per_second": 18965.63829978423,
+      "required_pages_per_second_upper_estimate": 0.0043873666368211385
     },
     {
       "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": 28455.456979235816,
-      "required_pages_per_second_upper_estimate": 0.0017327880357907582
+      "required_token_positions_per_second": 28448.457449676345,
+      "required_pages_per_second_upper_estimate": 0.00173236180116428
     },
     {
       "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": 28455.456979235816,
-      "required_pages_per_second_upper_estimate": 0.0021905798480079796
+      "required_token_positions_per_second": 28448.457449676345,
+      "required_pages_per_second_upper_estimate": 0.002190041004846554
     },
     {
       "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": 28455.456979235816,
-      "required_pages_per_second_upper_estimate": 0.35332154139384153
+      "required_token_positions_per_second": 28448.457449676345,
+      "required_pages_per_second_upper_estimate": 0.3532346306626314
     }
   ],
   "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": 0.08335933784357508,
-    "producer_p95_seconds": 0.3461581939991447,
-    "gpu_update_seconds": 0.9716238266767241,
-    "producer_batches_per_second": 11.996256518693965,
-    "gpu_batches_per_second": 1.0292048965290732,
-    "producer_headroom": 11.655848664489026,
+    "producer_batch_seconds": 0.10894301782042248,
+    "producer_p95_seconds": 0.36066934799964656,
+    "gpu_update_seconds": 0.9718628874309861,
+    "producer_batches_per_second": 9.179110511224886,
+    "gpu_batches_per_second": 1.0289517306740577,
+    "producer_headroom": 8.920836845487132,
     "sustainable": true,
-    "buffer_seconds": 60.240677253956896,
+    "buffer_seconds": 60.25549902072114,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369153.64899867366,
-    "upload_bytes_per_second": 5353222.851677012,
+    "generation_bytes_per_second": 369796.4607459037,
+    "upload_bytes_per_second": 5178901.939750174,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.21177961797864,
+    "estimated_full_shards_per_hour": 79.34971205504259,
     "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": 1.2097226180012512,
-      "bytes": 7368748,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.1227261329986504,
       "bytes": 5278941,
@@ -562,6 +556,12 @@
       "bytes": 10030549,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 2.951283398000669,
+      "bytes": 2617533,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -287,3 +287,4 @@
 {"step": 143700, "loss": 0.6280402727425098, "lr": 0.0003, "grad_norm": 0.15036053841949226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21030474349, "tokens_per_second": 149733.17884332212, "data_wait_s": 1.2964002962689847e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 143800, "loss": 0.5547901634126902, "lr": 0.0003, "grad_norm": 0.13474167818185723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21045143923, "tokens_per_second": 150466.8883624078, "data_wait_s": 1.4326000382425264e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 143900, "loss": 0.6152845062315464, "lr": 0.0003, "grad_norm": 0.14872422042671207, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21059769634, "tokens_per_second": 149212.66207554346, "data_wait_s": 1.4273002307163551e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 144000, "loss": 0.6275187488645315, "lr": 0.0003, "grad_norm": 0.14284842503990988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 21074305231, "tokens_per_second": 148443.2199237849, "data_wait_s": 1.4444998669205233e-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 sha256:8aac40bd9f7f3560399ebd631ed244353dff3e77ac5d57edf1b0c11e15f8596e
-size 616182
+oid sha256:e613e6b7bfdf7292abb041bfbed7a364054eb13ca2491e704785469d8d287d81
+size 618300