⚡ add humaneval support

Makefile

@@ -19,6 +19,13 @@ dev_install:
 	echo "Installing ktransformers"
 	KTRANSFORMERS_FORCE_BUILD=TRUE pip install -e . -v --no-build-isolation
 	echo "Installation completed successfully"
+clean:
+	rm -rf build
+	rm -rf *.egg-info
+	rm -rf ktransformers/ktransformers_ext/build
+	rm -rf ktransformers/ktransformers_ext/cuda/build
+	rm -rf ktransformers/ktransformers_ext/cuda/dist
+	rm -rf ktransformers/ktransformers_ext/cuda/*.egg-info	
 install_numa:
 	USE_NUMA=1 make dev_install
 install_no_numa:

ktransformers/tests/.gitignore

@@ -0,0 +1 @@
+humaneval/results

ktransformers/tests/humaneval/eval_api.py

@@ -0,0 +1,89 @@
+# adapt from https://github.com/abacaj/code-eval?tab=readme-ov-file
+import argparse
+import os
+import requests
+from human_eval.data import write_jsonl, read_problems
+import tqdm
+
+from evaluation import filter_code, fix_indents
+from prompts import instruct_prompt
+
+def generate_text(api_url,question , model_name, stream=False, auth_token=None):
+    headers = {
+        'accept': 'application/json',
+        'Content-Type': 'application/json',
+        # 添加 API Key
+        'Authorization' : 'Bearer ' + auth_token
+    }
+    question = instruct_prompt(question)
+    data = {
+        "messages": [{"content": question, "role": "user"}],
+        "model": model_name,
+        "stream": stream,
+        "temperature": 0.6
+    }
+    print(f"content: {question}")
+    response = requests.post(api_url, headers=headers, json=data,verify=False)
+    if response.status_code == 200:
+        result = response.json()
+        results = result.get('choices', [{}])[0].get('message', {}).get('content', '')
+        return [filter_code(fix_indents(results))]
+    else:
+        print(f"API Request failed with status code {response.status_code}")
+        return None
+
+def run_eval_api(
+    api_url: str,
+    model_name: str,
+    out_path: str,
+    format_tabs: bool = False,
+    auth_token: str = None,
+    problem_file: str = None,
+    append: bool = False
+):
+    if(problem_file is None):
+        problems = read_problems()
+    else:
+        problems = read_problems(problem_file)
+    samples = []
+    pbar = tqdm.tqdm(total=len(problems) * 1)
+    try:
+        for task_id in problems:
+            if format_tabs:
+                prompt = problems[task_id]["prompt"].replace("    ", "\t")
+            else:
+                prompt = problems[task_id]["prompt"]
+            completion = generate_text(api_url, prompt, model_name, auth_token=auth_token)
+            # samples.append({"task_id": task_id, "completion": completion})
+            for sample in completion:
+                result = dict(
+                    task_id=task_id,
+                    completion=sample,
+                )
+                samples += [result]
+                if append:
+                    write_jsonl(out_path, [result],append=append)
+            pbar.update(1)
+        if not append:
+            write_jsonl(out_path, samples,append=append)
+    except Exception as e:
+        write_jsonl(out_path, samples,append=append)
+        print(f"Error: {e}")
+
+def main(output_path, api_url, model_name, auth_token, format_tabs,problem_file, append):
+    os.makedirs(os.path.dirname(output_path), exist_ok=True)
+    run_eval_api(api_url, model_name, output_path, format_tabs, auth_token, problem_file,append)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="API Generate Tester")
+    parser.add_argument("--api_url", type=str, default="https://api.siliconflow.cn/v1/chat/completions", help="API URL")
+    parser.add_argument("--model_name", type=str, default="Pro/deepseek-ai/DeepSeek-V3", help="Model Name")
+    parser.add_argument("--out_path", type=str, default="results/api/eval.jsonl", help="Output Path")
+    parser.add_argument("--auth_token", type=str, default=None, help="Auth Token")
+    parser.add_argument("--format_tabs", action="store_true", help="Format Tabs")
+    parser.add_argument("--problem_file", type=str, default=None, help="Evalset File")
+    parser.add_argument("--no_append", action="store_false", help="Append to existing file")
+    args = parser.parse_args()
+    # api_url = "https://api.siliconflow.cn/v1/chat/completions"
+    main(args.out_path, args.api_url, args.model_name, args.auth_token, args.format_tabs, args.problem_file, args.no_append)

ktransformers/tests/humaneval/evaluation.py

@@ -0,0 +1,15 @@
+# reference: https://github.com/declare-lab/instruct-eval/blob/main/human_eval/main.py#L35
+def filter_code(completion: str) -> str:
+    # The program tends to overwrite, we only take the first function
+    completion = completion.lstrip("\n")
+    # we also remove ```python\n and ```
+    completion = completion.replace("```python\n", "").replace("```", "")
+    if 'if __name__ == "__main__":' in completion:
+        completion = completion.split('if __name__ == "__main__":')[0]
+    if "# Example usage" in completion:
+        completion = completion.split("# Example usage")[0]
+    return completion.split("\n\n")[0]
+
+
+def fix_indents(text: str) -> str:
+    return text.replace("\t", "    ")

ktransformers/tests/humaneval/prompts.py

@@ -0,0 +1,14 @@
+def instruct_prompt(prompt: str) -> str:
+    return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\nComplete the following Python code without any tests or explanation\n{prompt}\n\n### Response:"""
+
+
+def standard_prompt(prompt: str) -> str:
+    return f"""Complete the following Python code without any tests or explanation\n{prompt}"""
+
+
+def write_prompt(prompt: str) -> str:
+    return f"""Write a python program to complete the following code:\n{prompt}"""
+
+
+def replit_glaive_prompt(prompt: str) -> str:
+    return f"""Below is an instruction that describes a task, paired with an input that provides further context.\n Write a response that appropriately completes the request.\n\n ### Instruction:\nWrite a program to perform the given task.\n\n Input:\n{prompt}\n\n### Response:"""

ktransformers/tests/mmlu_pro_test.py

@@ -59,8 +59,8 @@ class DataEvaluator:
         :param text: The raw prediction string.
         :return: Processed prediction string.
         """
-        text = text.lstrip('\n').split('\n')[0]
-        return text[:1]
+        text = text.lstrip('\n').split('\n')[-1]
+        return text[-1:]
 
     def score(self, pred, answers):
         """

ktransformers/tests/mmlu_test.py

@@ -59,8 +59,8 @@ class DataEvaluator:
         :param text: The raw prediction string.
         :return: Processed prediction string.
         """
-        text = text.lstrip('\n').split('\n')[0]
-        return text[:1]
+        text = text.lstrip('\n').split('\n')[-1]
+        return text[-1:]
 
     def score(self, pred, answers):
         """