Auto-update results.csv

results.csv

@@ -11,7 +11,7 @@ RH SWE-Bench,rounakbende/rh-swe-bench,357,https://huggingface.co/datasets/rounak
 RH SWE-Bench,rounakbende/rh-swe-bench,357,https://huggingface.co/datasets/rounakbende/rh-swe-bench,OpenCode,[],True,https://github.com/opencode-ai/opencode,Qwen3.6-27B-FP8,RedHatAI/Qwen3.6-27B-FP8,True,27,fp8,https://huggingface.co/RedHatAI/Qwen3.6-27B-FP8,harbor,https://github.com/harbor-framework/harbor,357.0,10.0,0.44,410001939.0,0.0,2851571.0,412853510.0,115135.0,315451.0,42.22,1148464.0,0.0,7987.0,1156452.0,0.12,883.0,322.0
 RH SWE-Bench,rounakbende/rh-swe-bench,357,https://huggingface.co/datasets/rounakbende/rh-swe-bench,Pi,[],True,https://github.com/plandex-ai/plandex,Qwen3.6-27B-FP8,RedHatAI/Qwen3.6-27B-FP8,True,27,fp8,https://huggingface.co/RedHatAI/Qwen3.6-27B-FP8,harbor,https://github.com/harbor-framework/harbor,357.0,1.0,0.468,560672462.0,0.0,3618673.0,564291135.0,114791.0,333385.0,42.09,1570511.0,0.0,10136.0,1580647.0,0.12,933.0,321.0
 Shellbench,ShellBench/public-tasks,115,https://github.com/ShellBench/public-tasks,OpenClaw,[],True,https://github.com/OpenClaw/OpenClaw,Gemma4-31B-FP8,RedHatAI/gemma-4-31B-it-FP8-block,True,31,fp8,https://huggingface.co/RedHatAI/gemma-4-31B-it-FP8-block,harbor,https://github.com/harbor-framework/harbor,115.0,6.0,0.0,0.0,0.0,0.0,0.0,29895.0,36503.0,8.3,0.0,0.0,0.0,0.0,0.07,317.0,259.0
-Shellbench,ShellBench/public-tasks,115,https://github.com/ShellBench/public-tasks,OpenClaw,[],True,https://github.com/OpenClaw/OpenClaw,gpt-oss-120b,RedHatAI/gpt-oss-120b,True,120,,https://huggingface.co/RedHatAI/gpt-oss-120b,harbor,https://github.com/harbor-framework/harbor,115.0,1.0,0.0,0.0,0.0,0.0,0.0,8170.0,13296.0,2.27,0.0,0.0,0.0,0.0,0.02,115.0,71.0
+Shellbench,ShellBench/public-tasks,115,https://github.com/ShellBench/public-tasks,OpenClaw,[],True,https://github.com/OpenClaw/OpenClaw,GPT-OSS-120B,RedHatAI/gpt-oss-120b,True,120,,https://huggingface.co/RedHatAI/gpt-oss-120b,harbor,https://github.com/harbor-framework/harbor,115.0,1.0,0.0,0.0,0.0,0.0,0.0,8170.0,13296.0,2.27,0.0,0.0,0.0,0.0,0.02,115.0,71.0
 Shellbench,ShellBench/public-tasks,115,https://github.com/ShellBench/public-tasks,OpenClaw,[],True,https://github.com/OpenClaw/OpenClaw,Mistral-Small-4-119B-2603-NVFP4,RedHatAI/Mistral-Small-4-119B-2603-NVFP4,True,119,fp4,https://huggingface.co/RedHatAI/Mistral-Small-4-119B-2603-NVFP4,harbor,https://github.com/harbor-framework/harbor,115.0,2.0,0.009,0.0,0.0,0.0,0.0,4846.0,9862.0,1.35,0.0,0.0,0.0,0.0,0.01,85.0,42.0
 Shellbench,ShellBench/public-tasks,115,https://github.com/ShellBench/public-tasks,OpenClaw,[],True,https://github.com/OpenClaw/OpenClaw,Nemotron-3-Super-120B-NVFP4,RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,True,120,fp4,https://huggingface.co/RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4,harbor,https://github.com/harbor-framework/harbor,115.0,1.0,0.014,0.0,0.0,0.0,0.0,21575.0,26392.0,5.99,0.0,0.0,0.0,0.0,0.05,229.0,187.0
 Shellbench,ShellBench/public-tasks,115,https://github.com/ShellBench/public-tasks,OpenClaw,[],True,https://github.com/OpenClaw/OpenClaw,Qwen3.6-27B-FP8,RedHatAI/Qwen3.6-27B-FP8,True,27,fp8,https://huggingface.co/RedHatAI/Qwen3.6-27B-FP8,harbor,https://github.com/harbor-framework/harbor,115.0,1.0,0.128,0.0,0.0,0.0,0.0,10812.0,16949.0,6.01,0.0,0.0,0.0,0.0,0.05,147.0,94.0