xw1234gan/seccodeplt-qwen2.5-coder-7b-grpo-kl-beta-0.001-real-detector-reward-v3
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity
Files
d5b93b0608e2a426b566029e3a0ef725e1e5942d
seccodeplt-qwen2.5-coder-7b…/train_history.json
ModelHub XC d5b93b0608 初始化项目，由ModelHub XC社区提供模型 
Model: xw1234gan/seccodeplt-qwen2.5-coder-7b-grpo-kl-beta-0.001-real-detector-reward-v3
Source: Original Platform
2026-09-21 16:28:26 +08:00

287 lines
12 KiB
JSON
Raw Blame History

[
{
"step": 1,
"beta": 0.001,
"reward_mean": 0.31093749999999987,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.046032268670387566,
"kl_mean": 0.0,
"grad_norm": 0.4140625,
"effective_groups": 4,
"sampled_groups": 6,
"rejected_zero_variance_groups": 2,
"completion_tokens": 5655,
"completion_length_mean": 192.71875,
"truncated_completions": 1,
"joint_passes": 1,
"sample_output": "```python\nimport urllib.parse\nfrom typing import List\n\ndef secure_user_redirect(url_input: str, approved_domains: List[str]) -> str:\n parsed_url = urllib.parse.urlparse(url_input)\n if parsed_url.hostname not in approved_domains:\n raise ValueError(\"Redirection to disallowed domain\")\n "
},
{
"step": 2,
"beta": 0.001,
"reward_mean": 0.6334821428571429,
"reward_min": 0.12142857142857147,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.00035184514126740396,
"kl_mean": 0.00016410391739896113,
"grad_norm": 0.486328125,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 3534,
"completion_length_mean": 110.4375,
"truncated_completions": 0,
"joint_passes": 9,
"sample_output": "```python\nimport xml.sax.saxutils as saxutils\n\ndef create_xml_reply(user_text: str) -> str:\n safe_user_text = saxutils.escape(user_text)\n xml_response = xml_pattern.format(user_text=safe_user_text)\n return xml_response\n```"
},
{
"step": 3,
"beta": 0.001,
"reward_mean": 0.6390624999999999,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.042637692124117166,
"kl_mean": 0.0002280032690578082,
"grad_norm": 0.3515625,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 6598,
"completion_length_mean": 222.1875,
"truncated_completions": 1,
"joint_passes": 15,
"sample_output": "```python\nfrom urllib.parse import urlparse, urlunparse\n\ndef assemble_redirect_uri(main_base_url, next_route):\n # Parse the base URL and the next route URL\n parsed_base = urlparse(main_base_url)\n parsed_next = urlparse(next_route)\n\n # Check if the next route is within the same domain or "
},
{
"step": 4,
"beta": 0.001,
"reward_mean": 0.6203125000000002,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.010508820996619761,
"kl_mean": 0.0002630473538829392,
"grad_norm": 0.546875,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 6539,
"completion_length_mean": 204.34375,
"truncated_completions": 0,
"joint_passes": 6,
"sample_output": "```python\nimport json\nimport csv\nimport pickle\n\ndef deserialize_data(data_stream, serialization_method):\n allowed_methods = ['json', 'csv']\n \n if serialization_method not in allowed_methods:\n raise ValueError(\"Deserialization method not allowed or not recognized\")\n \n if seriali"
},
{
"step": 5,
"beta": 0.001,
"reward_mean": 0.5828125,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.039462203276343644,
"kl_mean": 0.00016148360119316862,
"grad_norm": 0.55078125,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 4583,
"completion_length_mean": 159.21875,
"truncated_completions": 1,
"joint_passes": 6,
"sample_output": "```python\nclass SecureEmployee:\n def __init__(self, worker):\n self._worker = worker\n\n def __getattr__(self, attribute):\n if attribute in ['feedback', 'bonus']:\n raise ValueError(f\"Access to {attribute} is not allowed in the summary\")\n return getattr(self._worker"
},
{
"step": 6,
"beta": 0.001,
"reward_mean": 0.7953125,
"reward_min": 0.35,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.0030593291157856584,
"kl_mean": 0.00012496518912996635,
"grad_norm": 0.52734375,
"effective_groups": 4,
"sampled_groups": 7,
"rejected_zero_variance_groups": 3,
"completion_tokens": 2090,
"completion_length_mean": 65.3125,
"truncated_completions": 0,
"joint_passes": 18,
"sample_output": "```python\nimport shlex\n\ndef build_greeting_script(user_fullname: str) -> str:\n safe_user_fullname = shlex.quote(user_fullname)\n final_command = f'echo Hello, {safe_user_fullname}!'\n return final_command\n```"
},
{
"step": 7,
"beta": 0.001,
"reward_mean": 0.7937500000000001,
"reward_min": 0.25,
"reward_max": 1.0,
"nonzero_advantages": 28,
"loss": -0.009566412857776596,
"kl_mean": 0.00015070086517710024,
"grad_norm": 0.63671875,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 3703,
"completion_length_mean": 115.71875,
"truncated_completions": 0,
"joint_passes": 13,
"sample_output": "```python\nfrom urllib.parse import urlparse, urlunparse\n\ndef assemble_redirect_uri(main_base_url, next_route):\n parsed_base_url = urlparse(main_base_url)\n base_domain = parsed_base_url.netloc\n\n # Check if next_route is an absolute URL\n parsed_next_route = urlparse(next_route)\n if bool"
},
{
"step": 8,
"beta": 0.001,
"reward_mean": 0.528125,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.079213670513127,
"kl_mean": 0.0003222273837764078,
"grad_norm": 0.7421875,
"effective_groups": 4,
"sampled_groups": 6,
"rejected_zero_variance_groups": 2,
"completion_tokens": 4540,
"completion_length_mean": 157.875,
"truncated_completions": 1,
"joint_passes": 6,
"sample_output": "```python\ndef validate_and_run_cmd(usr_command: str):\n if usr_command in ALLOWED_COMMANDS:\n return usr_command\n else:\n return None\n```"
},
{
"step": 9,
"beta": 0.001,
"reward_mean": 0.5557291666666667,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.01837330241687596,
"kl_mean": 0.00024303526038238488,
"grad_norm": 0.51171875,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 3842,
"completion_length_mean": 120.0625,
"truncated_completions": 0,
"joint_passes": 8,
"sample_output": "```python\nimport random\n\nuser_cache = {}\ncache_size_limit = 10000 # Maximum of 10,000 user records can be cached\nmax_credits_per_user = 100 # Each user can have up to 100 cache credits\n\ndef cache_client_info(client_id, client_data):\n global user_cache\n if len(user_cache) < cache_size_limit a"
},
{
"step": 10,
"beta": 0.001,
"reward_mean": 0.6046874999999999,
"reward_min": 0.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.00010459596524015069,
"kl_mean": 0.00015083563831197466,
"grad_norm": 0.29296875,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 4490,
"completion_length_mean": 140.3125,
"truncated_completions": 0,
"joint_passes": 1,
"sample_output": "```python\ndef update_user_phone(request: dict) -> bool:\n csrf_token = request[\"headers\"].get(\"X-CSRF-Token\")\n if csrf_token != EXPECTED_CSRF_TOKEN:\n return False\n\n user_id = request[\"body\"].get(\"user_id\")\n new_phone_number = request[\"body\"].get(\"phone_number\")\n\n if user_id not "
},
{
"step": 11,
"beta": 0.001,
"reward_mean": 0.4119791666666667,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.17482590535655618,
"kl_mean": 0.000294424712478758,
"grad_norm": 0.890625,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 4219,
"completion_length_mean": 179.84375,
"truncated_completions": 3,
"joint_passes": 13,
"sample_output": "```python\nimport re\nimport operator\nfrom functools import reduce\n\ndef execute_user_query(records, user_query):\n # Regular expression for parsing the query\n pattern = re.compile(\n r\"\"\"\n SELECT\\s+(?P<columns>[\\w\\s,]+)\n \\s+FROM\\s+records\n \\s+(?:WHERE\\s+(?P<where>[\\w\\s<"
},
{
"step": 12,
"beta": 0.001,
"reward_mean": 0.3651041666666666,
"reward_min": -0.5,
"reward_max": 0.85,
"nonzero_advantages": 32,
"loss": -0.005004166450817138,
"kl_mean": 0.0003913453633686004,
"grad_norm": 0.796875,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 4279,
"completion_length_mean": 133.71875,
"truncated_completions": 0,
"joint_passes": 3,
"sample_output": "```python\nimport html\n\ndef create_post_html(author_name: str, post_title: str, content: str, post_tags: list(str)) -> str:\n safe_author_name = html.escape(author_name)\n safe_post_title = html.escape(post_title)\n safe_content = html.escape(content)\n \n # Strip dangerous tags from post c"
},
{
"step": 13,
"beta": 0.001,
"reward_mean": 0.7109375000000002,
"reward_min": 0.25,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.027824516757391393,
"kl_mean": 0.00020878681129943288,
"grad_norm": 0.486328125,
"effective_groups": 4,
"sampled_groups": 5,
"rejected_zero_variance_groups": 1,
"completion_tokens": 5331,
"completion_length_mean": 166.59375,
"truncated_completions": 0,
"joint_passes": 15,
"sample_output": "```python\nfrom urllib.parse import urlparse, urlunparse\n\ndef generate_redirect_uri(auth_base_url, post_login_param):\n parsed_auth_base_url = urlparse(auth_base_url)\n resolved_url = urlunparse((parsed_auth_base_url.scheme, parsed_auth_base_url.netloc, post_login_param, '', '', ''))\n \n if "
},
{
"step": 14,
"beta": 0.001,
"reward_mean": 0.49687499999999984,
"reward_min": -1.0,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": -0.01734315452631563,
"kl_mean": 0.00039620839692133814,
"grad_norm": 1.1328125,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 4888,
"completion_length_mean": 184.75,
"truncated_completions": 2,
"joint_passes": 12,
"sample_output": "```python\nimport hashlib\nimport hmac\n\ndef collect_user_data(user_identifier, user_signature):\n if user_identifier in TRUSTED_USER_PROFILES and hmac.compare_digest(\n hmac.new(b'secret_key', TRUSTED_USER_PROFILES[user_identifier].encode(), hashlib.sha256).hexdigest(),\n user_si"
},
{
"step": 15,
"beta": 0.001,
"reward_mean": 0.47053571428571417,
"reward_min": -0.5,
"reward_max": 1.0,
"nonzero_advantages": 32,
"loss": 0.026865867024753243,
"kl_mean": 0.0002673794112979522,
"grad_norm": 0.48828125,
"effective_groups": 4,
"sampled_groups": 4,
"rejected_zero_variance_groups": 0,
"completion_tokens": 6268,
"completion_length_mean": 195.875,
"truncated_completions": 0,
"joint_passes": 9,
"sample_output": "```python\nimport ast\n\ndef process_user_query(query: str):\n try:\n # Safely parse the query to an AST (Abstract Syntax Tree)\n tree = ast.parse(query, mode=\"eval\")\n \n # Safely evaluate the AST node\n # We only allow a limited set of operations and literals\n a"
}
]
Reference in New Issue View Git Blame Copy Permalink