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": 74427.83944035564,
-      "required_pages_per_second_upper_estimate": 0.0064512164870555565
+      "required_token_positions_per_second": 73743.4898453045,
+      "required_pages_per_second_upper_estimate": 0.006391898798624715
     },
     {
       "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": 9303.479930044456,
-      "required_pages_per_second_upper_estimate": 0.00371129724351542
+      "required_token_positions_per_second": 9217.936230663063,
+      "required_pages_per_second_upper_estimate": 0.0036771725828399007
     },
     {
       "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": 18606.95986008891,
-      "required_pages_per_second_upper_estimate": 0.004304392692322566
+      "required_token_positions_per_second": 18435.872461326126,
+      "required_pages_per_second_upper_estimate": 0.004264814633654174
     },
     {
       "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": 27910.439790133365,
-      "required_pages_per_second_upper_estimate": 0.0016995993484586176
+      "required_token_positions_per_second": 27653.80869198919,
+      "required_pages_per_second_upper_estimate": 0.0016839718610209512
     },
     {
       "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": 27910.439790133365,
-      "required_pages_per_second_upper_estimate": 0.0021486229160867323
+      "required_token_positions_per_second": 27653.80869198919,
+      "required_pages_per_second_upper_estimate": 0.0021288667437512464
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27910.439790133365,
-      "required_pages_per_second_upper_estimate": 0.346554251960383
+      "required_token_positions_per_second": 27653.80869198919,
+      "required_pages_per_second_upper_estimate": 0.3433677526104671
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13148147678901978,
-    "producer_p95_seconds": 0.4053869780036621,
-    "gpu_update_seconds": 0.9905970743525818,
-    "producer_batches_per_second": 7.605634074255481,
-    "gpu_batches_per_second": 1.0094921799093377,
-    "producer_headroom": 7.534118862553787,
+    "producer_batch_seconds": 0.15260623889878389,
+    "producer_p95_seconds": 0.40561408800567733,
+    "gpu_update_seconds": 0.9997899496574274,
+    "producer_batches_per_second": 6.552812042391335,
+    "gpu_batches_per_second": 1.0002100944729886,
+    "producer_headroom": 6.551435621977016,
     "sustainable": true,
-    "buffer_seconds": 60.426421535507494,
+    "buffer_seconds": 60.987186929103075,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409732.17670146807,
-    "upload_bytes_per_second": 3795305.4286950855,
+    "generation_bytes_per_second": 430603.0380835422,
+    "upload_bytes_per_second": 3756505.766093821,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.91898704321892,
+    "estimated_full_shards_per_hour": 92.39738804702472,
     "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": 0.9035562749995734,
-      "bytes": 6454400,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.2266940429981332,
       "bytes": 2571680,
@@ -562,6 +556,12 @@
       "bytes": 6161152,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9979428110018489,
+      "bytes": 2470155,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -364,3 +364,4 @@
 {"step": 36400, "loss": 0.6693796012550592, "lr": 0.0003, "grad_norm": 0.15807813006973326, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5326521022, "tokens_per_second": 147688.64795002664, "data_wait_s": 1.2843003787565976e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 36500, "loss": 0.6346561908721924, "lr": 0.0003, "grad_norm": 0.14148785048075094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5341172877, "tokens_per_second": 149755.7742678632, "data_wait_s": 1.452199649065733e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 36600, "loss": 0.63532455265522, "lr": 0.0003, "grad_norm": 0.14360385765486508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5355806307, "tokens_per_second": 148635.82236582195, "data_wait_s": 1.1292999261058867e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 36700, "loss": 0.6649079211056232, "lr": 0.0003, "grad_norm": 0.1398867633620141, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5370553146, "tokens_per_second": 149866.2977470326, "data_wait_s": 8.098002581391484e-06, "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:aae9f094d4ffa87822cfb87caae9b3bca06506a5d2f9ff531eec77aa93ee9a04
-size 775060
+oid sha256:fb098976202dde221cb563b1b5517cb49e36f183d92f043ec66a84e97bc553fd
+size 777178