Update README.md

README.md

@@ -95,6 +95,17 @@ Both gpt-oss models can be fine-tuned for a variety of specialized use cases.
 - Do not use this model for creating nuclear, biological and chemical weapons.
 - Do not allow harmful or malicious outputs
 
+Code to duplicate the benchmark (Using +std for final result)
+```py
+
+#gpqa diamond
+!lm_eval --model hf --model_args pretrained=EpistemeAI/metatune-gpt20b-R1.2,parallelize=True,dtype=bfloat16 --tasks gpqa_diamond_cot_zeroshot  --num_fewshot 0 --gen_kwargs temperature=0.9,top_p=0.9,max_new_tokens=2048 --batch_size auto:4 --limit 10  --device cuda:0 --output_path ./eval_harness/gpt-oss-20b3
+#gsm8k cot
+!lm_eval --model hf --model_args pretrained=EpistemeAI/metatune-gpt20b-R1.2,parallelize=True,dtype=bfloat16 --tasks gsm8k_cot_llama --apply_chat_template --fewshot_as_multiturn  --num_fewshot 0 --gen_kwargs temperature=0.9,top_p=0.9,max_new_tokens=1024 --batch_size auto:4 --limit 10  --device cuda:0 --output_path ./eval_harness/gpt-oss-20b3
+#mmlu computer science
+!lm_eval --model hf --model_args pretrained=EpistemeAI/metatune-gpt20b-R1.2,parallelize=True,dtype=bfloat16 --tasks mmlu_pro_plus_computer_science --apply_chat_template --fewshot_as_multiturn  --num_fewshot 0 --gen_kwargs temperature=0.9,top_p=0.9,max_new_tokens=1024 --batch_size auto:4 --limit 10  --device cuda:0 --output_path ./eval_harness/gpt-oss-20b3
+
+```
 
 ## Benchmark
 hf (pretrained=EpistemeAI/metatune-gpt20b-R1.1,parallelize=True,dtype=bfloat16), gen_kwargs: (temperature=0.9,top_p=0.9,max_new_tokens=2048), limit: 10.0, num_fewshot: 0, batch_size: auto:4