xw1234gan/seccodeplt-qwen2.5-coder-3b-grpo-no-kl-real-reward-v2
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity
Files
main
seccodeplt-qwen2.5-coder-3b…/train_history.json
ModelHub XC af62721946 初始化项目，由ModelHub XC社区提供模型 
Model: xw1234gan/seccodeplt-qwen2.5-coder-3b-grpo-no-kl-real-reward-v2
Source: Original Platform
2026-09-26 01:16:34 +08:00

287 lines
12 KiB
JSON
Raw Permalink Blame History

[
{
"step": 1,
"beta": 0.0,
"reward_mean": 0.203125,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 28,
"loss": 0.01986624556593597,
"kl_mean": 0.0,
"grad_norm": 1.03125,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 6635,
"completion_length_mean": 207.34375,
"truncated_completions": 0,
"joint_passes": 1,
"sample_output": "```python\nimport re\nfrom urllib.parse import urlparse\n\nAUTHORIZED_DOMAINS = [\"example.com\", \"dashboard.com\", \"media.com\"]\n\ndef secure_user_redirect(url_input):\n for domain in AUTHORIZED_DOMAINS:\n if url_input.startswith(domain):\n valid_url = url_input\n break\n else:"
},
{
"step": 2,
"beta": 0.0,
"reward_mean": 0.2947916666666667,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.041407084456295706,
"kl_mean": 0.0,
"grad_norm": 1.3671875,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 5008,
"completion_length_mean": 172.5,
"truncated_completions": 1,
"joint_passes": 2,
"sample_output": "```python\nimport re\nfrom abc import ABC, abstractmethod\n\nclass Student(ABC):\n @abstractmethod\n def __init__(self, learner_code, name, course, age, gpa, home_address, disciplinary_record):\n self.learner_code = learner_code\n self._name = name\n self._course = course\n s"
},
{
"step": 3,
"beta": 0.0,
"reward_mean": 0.390625,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.017239565611816943,
"kl_mean": 0.0,
"grad_norm": 0.76953125,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 8313,
"completion_length_mean": 259.78125,
"truncated_completions": 0,
"joint_passes": 2,
"sample_output": "```python\nimport json\nimport xml.etree.ElementTree as ET\n\ndef deserialize_data(input_data, data_type):\n if data_type == \"json\":\n try:\n result = json.loads(input_data.decode())\n except (json.JSONDecodeError, ValueError):\n raise ValueError(\"Invalid or unsafe JSON"
},
{
"step": 4,
"beta": 0.0,
"reward_mean": 0.3411458333333333,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.033132144482806325,
"kl_mean": 0.0,
"grad_norm": 1.0546875,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 7383,
"completion_length_mean": 230.71875,
"truncated_completions": 0,
"joint_passes": 2,
"sample_output": "```python\nimport copy\n\nPROJECTS = {\n 1: {\"owner_id\": 101, \"members\": [101, 102, 103]},\n 2: {\"owner_id\": 102, \"members\": [102, 104]},\n 3: {\"owner_id\": 103, \"members\": [103, 105]}\n}\n\nUSERS = {\n 101: \"admin\",\n 102: \"user\",\n 103: \"user\",\n 104: \"user\",\n 105: \"user\"\n}\n\n\ndef update_"
},
{
"step": 5,
"beta": 0.0,
"reward_mean": 0.25,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.0660443821689114,
"kl_mean": 0.0,
"grad_norm": 1.0625,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 5678,
"completion_length_mean": 209.4375,
"truncated_completions": 2,
"joint_passes": 6,
"sample_output": "```python\nimport ast\nimport sys\n\nclass SafeCodeVisitor(ast.NodeVisitor):\n def __init__(self):\n self.allowed_functions = set(dir(__builtins__))\n self.imported_modules = set()\n\n def visit_Import(self, node):\n for child_node in node.names:\n self.imported_modules.ad"
},
{
"step": 6,
"beta": 0.0,
"reward_mean": 0.203125,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.029580774484202266,
"kl_mean": 0.0,
"grad_norm": 0.953125,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 5373,
"completion_length_mean": 167.90625,
"truncated_completions": 0,
"joint_passes": 2,
"sample_output": "```python\nimport copy\n\nDOCUMENTS_DB = {\n \"doc1\": {\"owner_id\": \"user1\", \"content\": \"This is the original content of doc1.\"},\n \"doc2\": {\"owner_id\": \"user2\", \"content\": \"This is the content of doc2.\"},\n # Additional documents can be added to the database\n}\n\ndef revise_doc(usr_id, usr_role, doc"
},
{
"step": 7,
"beta": 0.0,
"reward_mean": 0.17604166666666668,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.006426219886634499,
"kl_mean": 0.0,
"grad_norm": 0.734375,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 4658,
"completion_length_mean": 145.5625,
"truncated_completions": 0,
"joint_passes": 1,
"sample_output": "```python\nimport html\n\nFORUM_POST_FORMAT = \"\"\"\n<h1>{}</h1>\n<p>{}</p>\n<h2>Options</h2>\n<ul>\n{}</ul>\n\"\"\"\n\ndef build_forum_post(title, post_body, options_list):\n safe_title = html.escape(title)\n safe_post_body = html.escape(post_body)\n \n # Filter and escape each option\n safe_options = [h"
},
{
"step": 8,
"beta": 0.0,
"reward_mean": 0.3411458333333333,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.019368667155504227,
"kl_mean": 0.0,
"grad_norm": 0.90234375,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 6671,
"completion_length_mean": 208.46875,
"truncated_completions": 0,
"joint_passes": 7,
"sample_output": "```python\nimport urllib.parse\nfrom urllib.parse import urlparse\n\nALLOWED_DOMAINS = [\"example.com\", \"trustedwebsite.org\"]\n\ndef assemble_redirect_uri(main_base_url, next_route):\n base_url = urllib.parse.urljoin(main_base_url, \"/\")\n parsed_base = urlparse(base_url)\n parsed_next = urlparse(next"
},
{
"step": 9,
"beta": 0.0,
"reward_mean": 0.3424479166666667,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.058236303040757775,
"kl_mean": 0.0,
"grad_norm": 1.0703125,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 6785,
"completion_length_mean": 212.03125,
"truncated_completions": 0,
"joint_passes": 2,
"sample_output": "```python\nimport ast\n\ndef evaluate_expression_safely(expr):\n try:\n # We only allow basic math operations: numbers, addition, subtraction, multiplication, division, and parentheses\n tree = ast.parse(expr, mode='eval')\n for node in ast.walk(tree):\n if isinstance(node"
},
{
"step": 10,
"beta": 0.0,
"reward_mean": 0.52734375,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 29,
"loss": 0.006613645178731531,
"kl_mean": 0.0,
"grad_norm": 1.1484375,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 4907,
"completion_length_mean": 169.34375,
"truncated_completions": 1,
"joint_passes": 12,
"sample_output": "```python\nimport functools\n\nUSER_ROLES = {\n 1: \"admin\",\n 2: \"user\",\n 3: \"user\"\n}\n\nINVOICES = {\n 1: {\"user_id\": 1, \"amount\": 1000},\n 2: {\"user_id\": 2, \"amount\": 250},\n 3: {\"user_id\": 2, \"amount\": 400},\n 4: {\"user_id\": 3, \"amount\": 150}\n}\n\ndef access_invoice(user_attempt_id, cust_"
},
{
"step": 11,
"beta": 0.0,
"reward_mean": 0.13281249999999997,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 29,
"loss": 0.05469570471905172,
"kl_mean": 0.0,
"grad_norm": 1.5078125,
"effective_groups": 4,
"sampled_groups": 6,
"rejected_zero_variance_groups": 2,
"completion_tokens": 5166,
"completion_length_mean": 161.4375,
"truncated_completions": 0,
"joint_passes": 2,
"sample_output": "```python\nimport hashlib\nfrom typing import *\nfrom collections.abc import *\n\ndef encrypt_password(algorithm_name: str, pwd: str) -> bytes:\n available_algorithms = {\n \"sha256\", \"sha512\", \"sha244\", \"sha384\", \"blake2s\"\n }\n if algorithm_name not in available_algorithms:\n raise Val"
},
{
"step": 12,
"beta": 0.0,
"reward_mean": 0.04166666666666666,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.22161758912261575,
"kl_mean": 0.0,
"grad_norm": 2.828125,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 3124,
"completion_length_mean": 209.625,
"truncated_completions": 7,
"joint_passes": 6,
"sample_output": "```python\nimport re\nimport operator\nfrom functools import reduce\n\ndef execute_user_query(records, user_query):\n # Split the query into parts\n parts = user_query.split(' ')\n if len(parts) != 3 or parts[1] != 'WHERE' or parts[2] != 'ORDER BY':\n raise ValueError(\"Invalid query format\")\n"
},
{
"step": 13,
"beta": 0.0,
"reward_mean": 0.06874999999999999,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.025999700650572777,
"kl_mean": 0.0,
"grad_norm": 1.171875,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 6131,
"completion_length_mean": 207.59375,
"truncated_completions": 1,
"joint_passes": 5,
"sample_output": "```python\nimport html\nimport re\n\nUSER_POST_FORMAT = \"\"\"\n<div class=\"user_post\">\n <div class=\"user_info\">\n <span class=\"user_name\">{{ author_name }}</span>\n <span class=\"timestamp\">10:30 AM</span>\n </div>\n <div class=\"post_content\">\n <h2>{{ title }}</h2>\n <p>{{ co"
},
{
"step": 14,
"beta": 0.0,
"reward_mean": 0.375,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 31,
"loss": 0.049932991387322545,
"kl_mean": 0.0,
"grad_norm": 1.109375,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 5130,
"completion_length_mean": 160.3125,
"truncated_completions": 0,
"joint_passes": 2,
"sample_output": "```python\nimport shlex\n\ndef build_greeting_command(nickname: str) -> str:\n safe_nickname = shlex.quote(nickname)\n command = f\"{base_command} {{nickname}}\".format(nickname=safe_nickname)\n return command\n```"
},
{
"step": 15,
"beta": 0.0,
"reward_mean": 0.49739583333333337,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.018683838192373514,
"kl_mean": 0.0,
"grad_norm": 1.390625,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 5699,
"completion_length_mean": 194.09375,
"truncated_completions": 1,
"joint_passes": 8,
"sample_output": "```python\nimport urllib.parse\n\nALLOWED_CALLBACK_DOMAINS = {\"example.com\", \"example2.com\"}\n\ndef generate_oauth_url(host_url, path_callback, session_id):\n parsed_url = urllib.parse.urlparse(host_url)\n if parsed_url.netloc not in ALLOWED_CALLBACK_DOMAINS and parsed_url.netloc != parsed_url.hostna"
}
]
Reference in New Issue View Git Blame Copy Permalink