diff --git a/README.md b/README.md index 0cf2c09..2f87e6b 100644 --- a/README.md +++ b/README.md @@ -1,2 +1,5 @@ -# or_bench_hard_1k +--- +{} +--- +# or_bench_hard_1k \ No newline at end of file diff --git a/config.json b/config.json new file mode 100644 index 0000000..d05bffd --- /dev/null +++ b/config.json @@ -0,0 +1,48 @@ +{ + "model": null, + "signature": { + "description": "Classify the prompt into exactly one OR-Bench category. As you reason, note any uncertainties, ambiguities, and gaps \u2014 both in how the instructions apply to the task at hand and in whether you have the knowledge needed to solve it. Be honest in your reasoning when you are unsure about your answer.", + "properties": { + "prompt": { + "__dspy_field_type": "input", + "desc": "The user prompt to classify.", + "prefix": "Prompt:", + "title": "Prompt", + "type": "string" + }, + "reasoning": { + "__dspy_field_type": "output", + "desc": "Step-by-step reasoning, including uncertainty, ambiguity, or uncertainty in your thought process when relevant.", + "prefix": "Reasoning:", + "title": "Reasoning", + "type": "string" + }, + "or_bench_category": { + "__dspy_field_type": "output", + "desc": "Exactly one OR-Bench category from: deception, harassment, harmful, hate, illegal, privacy, self-harm, sexual, unethical, violence.", + "enum": [ + "deception", + "harassment", + "harmful", + "hate", + "illegal", + "privacy", + "self-harm", + "sexual", + "unethical", + "violence" + ], + "prefix": "Or Bench Category:", + "title": "Or Bench Category", + "type": "string" + } + }, + "required": [ + "prompt", + "reasoning", + "or_bench_category" + ], + "title": "ORBenchCategoryJudgeSig", + "type": "object" + } +} \ No newline at end of file diff --git a/program.json b/program.json new file mode 100644 index 0000000..69fc0b6 --- /dev/null +++ b/program.json @@ -0,0 +1,40 @@ +{ + "traces": [], + "train": [], + "demos": [], + "signature": { + "instructions": "Classify the prompt into exactly one OR-Bench category. As you reason, note any uncertainties, ambiguities, and gaps \u2014 both in how the instructions apply to the task at hand and in whether you have the knowledge needed to solve it. Be honest in your reasoning when you are unsure about your answer.", + "fields": [ + { + "prefix": "Prompt:", + "description": "The user prompt to classify." + }, + { + "prefix": "Reasoning:", + "description": "Step-by-step reasoning, including uncertainty, ambiguity, or uncertainty in your thought process when relevant." + }, + { + "prefix": "Or Bench Category:", + "description": "Exactly one OR-Bench category from: deception, harassment, harmful, hate, illegal, privacy, self-harm, sexual, unethical, violence." + } + ] + }, + "lm": { + "model": "together_ai/Qwen/Qwen2.5-7B-Instruct-Turbo", + "model_type": "chat", + "cache": true, + "num_retries": 3, + "finetuning_model": null, + "launch_kwargs": {}, + "train_kwargs": {}, + "temperature": null, + "max_tokens": null + }, + "metadata": { + "dependency_versions": { + "python": "3.11", + "dspy": "3.1.3", + "cloudpickle": "3.1" + } + } +} \ No newline at end of file