NecroMOnk/malicious-coding-intent-v6

1{2  "model_dir": "models\\v6_code_aware_50k_oss_clean_benign_code",3  "holdout": "data\\clf\\benign_code_holdout_github_lora_clean.jsonl",4  "overall": {5    "n": 12000,6    "threshold": 0.5,7    "false_positive_rate": 0.0092,8    "flagged": 110,9    "score_mean": 0.022043,10    "score_p50": 0.000761,11    "score_p90": 0.037042,12    "score_p95": 0.104331,13    "score_p99": 0.469681,14    "score_max": 0.99435815  },16  "by_source": {17    "local_project_code": {18      "n": 160,19      "threshold": 0.5,20      "false_positive_rate": 0.0063,21      "flagged": 1,22      "score_mean": 0.027283,23      "score_p50": 0.002633,24      "score_p90": 0.034741,25      "score_p95": 0.107271,26      "score_p99": 0.437606,27      "score_max": 0.95558228    },29    "local_repo_hs": {30      "n": 14,31      "threshold": 0.5,32      "false_positive_rate": 0.1429,33      "flagged": 2,34      "score_mean": 0.234466,35      "score_p50": 0.173636,36      "score_p90": 0.50088,37      "score_p95": 0.601973,38      "score_p99": 0.732286,39      "score_max": 0.76486440    },41    "local_repo_isre": {42      "n": 173,43      "threshold": 0.5,44      "false_positive_rate": 0.0,45      "flagged": 0,46      "score_mean": 0.015629,47      "score_p50": 0.000923,48      "score_p90": 0.065123,49      "score_p95": 0.111808,50      "score_p99": 0.148813,51      "score_max": 0.17590952    },53    "local_repo_job_application_pipeline": {54      "n": 444,55      "threshold": 0.5,56      "false_positive_rate": 0.0023,57      "flagged": 1,58      "score_mean": 0.016508,59      "score_p50": 0.001555,60      "score_p90": 0.038672,61      "score_p95": 0.117372,62      "score_p99": 0.200662,63      "score_max": 0.69193464    },65    "local_repo_llama_cpp": {66      "n": 1000,67      "threshold": 0.5,68      "false_positive_rate": 0.023,69      "flagged": 23,70      "score_mean": 0.050298,71      "score_p50": 0.005734,72      "score_p90": 0.128486,73      "score_p95": 0.259195,74      "score_p99": 0.691108,75      "score_max": 0.99182476    },77    "local_repo_olympiad_math": {78      "n": 53,79      "threshold": 0.5,80      "false_positive_rate": 0.0189,81      "flagged": 1,82      "score_mean": 0.032375,83      "score_p50": 0.001522,84      "score_p90": 0.075568,85      "score_p95": 0.202259,86      "score_p99": 0.444325,87      "score_max": 0.56193388    },89    "local_repo_packing": {90      "n": 316,91      "threshold": 0.5,92      "false_positive_rate": 0.0032,93      "flagged": 1,94      "score_mean": 0.013281,95      "score_p50": 0.000252,96      "score_p90": 0.008419,97      "score_p95": 0.049662,98      "score_p99": 0.323688,99      "score_max": 0.71564100    },101    "local_repo_pipeline": {102      "n": 136,103      "threshold": 0.5,104      "false_positive_rate": 0.0074,105      "flagged": 1,106      "score_mean": 0.021312,107      "score_p50": 0.001563,108      "score_p90": 0.037614,109      "score_p95": 0.080462,110      "score_p99": 0.369778,111      "score_max": 0.888886112    },113    "local_repo_repo": {114      "n": 13,115      "threshold": 0.5,116      "false_positive_rate": 0.0,117      "flagged": 0,118      "score_mean": 0.049554,119      "score_p50": 0.003904,120      "score_p90": 0.15528,121      "score_p95": 0.195039,122      "score_p99": 0.231913,123      "score_max": 0.241131124    },125    "local_repo_utils": {126      "n": 114,127      "threshold": 0.5,128      "false_positive_rate": 0.0088,129      "flagged": 1,130      "score_mean": 0.017779,131      "score_p50": 0.000741,132      "score_p90": 0.021215,133      "score_p95": 0.039684,134      "score_p99": 0.316717,135      "score_max": 0.970841136    },137    "local_repo_vesuvius": {138      "n": 730,139      "threshold": 0.5,140      "false_positive_rate": 0.0548,141      "flagged": 40,142      "score_mean": 0.103365,143      "score_p50": 0.018504,144      "score_p90": 0.349818,145      "score_p95": 0.531822,146      "score_p99": 0.915992,147      "score_max": 0.994358148    },149    "python_stdlib": {150      "n": 8847,151      "threshold": 0.5,152      "false_positive_rate": 0.0044,153      "flagged": 39,154      "score_mean": 0.012388,155      "score_p50": 0.000454,156      "score_p90": 0.017894,157      "score_p95": 0.047606,158      "score_p99": 0.258983,159      "score_max": 0.97892160    }161  },162  "flagged_examples": [163    {164      "score": 0.955582,165      "source": "local_project_code",166      "path": "C:\\GitHub\\Safety DS\\scripts\\build_malware_code_pool.py",167      "preview": "def download_vxunderground( spec: dict, builder: PoolBuilder, chunk_cfg: dict, insecure: bool ) -> None: repo = spec[\"repo\"] cache = ROOT / spec.get(\"cache_dir\", \"data/external/vxunderground\") cache.mkdir(parents=True, exist_ok=True) for su"168    },169    {170      "score": 0.71564,171      "source": "local_repo_packing",172      "path": "C:\\GitHub\\packing\\core\\pack_cuda_primitives.py",173      "preview": "ss) forward_counts[0] = n_subj; for (int i = 0; i < n_subj; ++i) { forward_polys[0][i] = subj[i]; } } else { // Initialize current buffer current_count = n_subj; for (int i = 0; i < n_subj; ++i) { current_poly[i] = subj[i]; } } // Apply eac"174    },175    {176      "score": 0.586343,177      "source": "local_repo_llama_cpp",178      "path": "C:\\lora_training\\llama.cpp\\convert_hf_to_gguf.py",179      "preview": "def __init__(self, *args, **kwargs): super().__init__(*args, **kwargs) # SwigLU activation assert self.hparams[\"activation_function\"] == \"swiglu\" # ALiBi position embedding assert self.hparams[\"position_embedding_type\"] == \"alibi\" # Embeddi"180    },181    {182      "score": 0.863325,183      "source": "local_repo_llama_cpp",184      "path": "C:\\lora_training\\llama.cpp\\common\\arg.cpp",185      "preview": "_ARG_NO_KV_OFFLOAD\")); add_opt(common_arg( {\"-nr\", \"--no-repack\"}, \"disable weight repacking\", [](common_params & params) { params.no_extra_bufts = true; } ).set_env(\"LLAMA_ARG_NO_REPACK\")); add_opt(common_arg( {\"--no-host\"}, \"bypass host b"186    },187    {188      "score": 0.991824,189      "source": "local_repo_llama_cpp",190      "path": "C:\\lora_training\\llama.cpp\\common\\arg.cpp",191      "preview": "nd_dev_t> devices; for (size_t i = 0; i < ggml_backend_dev_count(); ++i) { auto * dev = ggml_backend_dev_get(i); if (ggml_backend_dev_type(dev) != GGML_BACKEND_DEVICE_TYPE_CPU) { devices.push_back(dev); } } printf(\"Available devices:\\n\"); f"192    },193    {194      "score": 0.60719,195      "source": "local_repo_llama_cpp",196      "path": "C:\\lora_training\\llama.cpp\\common\\arg.cpp",197      "preview": "n_ubatch = 1024; params.n_batch = 1024; params.n_ctx = 0; params.n_cache_reuse = 256; } ).set_examples({LLAMA_EXAMPLE_SERVER})); add_opt(common_arg( {\"--fim-qwen-7b-spec\"}, string_format(\"use Qwen 2.5 Coder 7B + 0.5B draft for speculative d"198    },199    {200      "score": 0.827293,201      "source": "local_repo_llama_cpp",202      "path": "C:\\lora_training\\llama.cpp\\common\\arg.cpp",203      "preview": "wen 3 Coder 30B A3B Instruct (note: can download weights from the internet)\"), [](common_params & params) { params.model.hf_repo = \"ggml-org/Qwen3-Coder-30B-A3B-Instruct-Q8_0-GGUF\"; params.model.hf_file = \"qwen3-coder-30b-a3b-instruct-q8_0."204    },205    {206      "score": 0.534571,207      "source": "local_repo_llama_cpp",208      "path": "C:\\lora_training\\llama.cpp\\common\\base64.hpp",209      "preview": "; return 62; } else if (c == '_') { alphabet = alphabet::url_filename_safe; return 63; } } throw base64_error(\"invalid base64 character.\"); } }; #endif // !PUBLIC_DOMAIN_BASE64_HPP_"210    },211    {212      "score": 0.551723,213      "source": "local_repo_llama_cpp",214      "path": "C:\\lora_training\\llama.cpp\\common\\chat-parser.cpp",215      "preview": "n_regex preamble_regex(\"<\\\\|channel\\\\|>commentary\"); static const common_regex tool_call1_regex(recipient + \"<\\\\|channel\\\\|>(analysis|commentary)\" + constraint + \"?\"); static const common_regex tool_call2_regex(\"<\\\\|channel\\\\|>(analysis|com"216    },217    {218      "score": 0.536037,219      "source": "local_repo_llama_cpp",220      "path": "C:\\lora_training\\llama.cpp\\common\\chat-parser.cpp",221      "preview": "case COMMON_CHAT_FORMAT_DEEPSEEK_R1: common_chat_parse_deepseek_r1(builder); break; case COMMON_CHAT_FORMAT_DEEPSEEK_V3_1: common_chat_parse_deepseek_v3_1(builder); break; case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2: common_chat_parse_function"222    },223    {224      "score": 0.616117,225      "source": "local_repo_llama_cpp",226      "path": "C:\\lora_training\\llama.cpp\\common\\chat-parser.cpp",227      "preview": "common_chat_parse_kimi_k2(builder); break; case COMMON_CHAT_FORMAT_QWEN3_CODER_XML: common_chat_parse_qwen3_coder_xml(builder); break; case COMMON_CHAT_FORMAT_APRIEL_1_5: common_chat_parse_apriel_1_5(builder); break; case COMMON_CHAT_FORMAT"228    },229    {230      "score": 0.609806,231      "source": "local_repo_llama_cpp",232      "path": "C:\\lora_training\\llama.cpp\\common\\chat.cpp",233      "preview": "msg_new.tool_calls.size() < msg_prv.tool_calls.size()) { throw std::runtime_error(\"Invalid diff: now finding less tool calls!\"); } if (!msg_prv.tool_calls.empty()) { const auto idx = msg_prv.tool_calls.size() - 1; const auto & pref = msg_pr"234    },235    {236      "score": 0.876555,237      "source": "local_repo_llama_cpp",238      "path": "C:\\lora_training\\llama.cpp\\common\\chat.cpp",239      "preview": "y v3.1 Llama 3.1\"; case COMMON_CHAT_FORMAT_DEEPSEEK_V3_1: return \"DeepSeek V3.1\"; case COMMON_CHAT_FORMAT_HERMES_2_PRO: return \"Hermes 2 Pro\"; case COMMON_CHAT_FORMAT_COMMAND_R7B: return \"Command R7B\"; case COMMON_CHAT_FORMAT_GRANITE: retur"240    },241    {242      "score": 0.610185,243      "source": "local_repo_llama_cpp",244      "path": "C:\\lora_training\\llama.cpp\\common\\chat.h",245      "preview": "OMMON_CHAT_FORMAT_GRANITE, COMMON_CHAT_FORMAT_GPT_OSS, COMMON_CHAT_FORMAT_SEED_OSS, COMMON_CHAT_FORMAT_NEMOTRON_V2, COMMON_CHAT_FORMAT_APERTUS, COMMON_CHAT_FORMAT_LFM2_WITH_JSON_TOOLS, COMMON_CHAT_FORMAT_GLM_4_5, COMMON_CHAT_FORMAT_MINIMAX_"246    },247    {248      "score": 0.864733,249      "source": "local_repo_llama_cpp",250      "path": "C:\\lora_training\\llama.cpp\\common\\common.cpp",251      "preview": "d-%H_%M_%S\", std::localtime(&as_time_t)); const int64_t ns = std::chrono::duration_cast<std::chrono::nanoseconds>( current_time.time_since_epoch() % 1000000000).count(); char timestamp_ns[11]; snprintf(timestamp_ns, 11, \"%09\" PRId64, ns); r"252    },253    {254      "score": 0.972733,255      "source": "local_repo_llama_cpp",256      "path": "C:\\lora_training\\llama.cpp\\common\\common.cpp",257      "preview": "ARATOR; } return p; }; if (getenv(\"LLAMA_CACHE\")) { cache_directory = std::getenv(\"LLAMA_CACHE\"); } else { #if defined(__linux__) || defined(__FreeBSD__) || defined(_AIX) || defined(__OpenBSD__) if (std::getenv(\"XDG_CACHE_HOME\")) { cache_di"258    },259    {260      "score": 0.526275,261      "source": "local_repo_llama_cpp",262      "path": "C:\\lora_training\\llama.cpp\\common\\common.cpp",263      "preview": "mmon_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_MIN_P); get_float(llama_model_meta_key_str(LLAMA_MODEL_META_KEY_SAMPLING_XTC_PROBABILITY), sparams.xtc_probability, common_params_sampling_config::COMMON_PARAMS_SAMPLING_CONFIG_XTC_"264    },265    {266      "score": 0.539756,267      "source": "local_repo_llama_cpp",268      "path": "C:\\lora_training\\llama.cpp\\common\\download.cpp",269      "preview": "l_successful = common_pull_file(cli, parts.path, path_temporary, supports_ranges, existing_size, total_size); if (!was_pull_successful) { if (i + 1 < max_attempts) { const int exponential_backoff_delay = std::pow(retry_delay_seconds, i) * 1"270    },271    {272      "score": 0.530205,273      "source": "local_repo_llama_cpp",274      "path": "C:\\lora_training\\llama.cpp\\common\\download.cpp",275      "preview": "size_t len) { buf.insert(buf.end(), data, data + len); return params.max_size == 0 || buf.size() <= static_cast<size_t>(params.max_size); }, nullptr ); if (!res) { throw std::runtime_error(\"error: cannot make GET request\"); } return { res->"276    },277    {278      "score": 0.882664,279      "source": "local_repo_llama_cpp",280      "path": "C:\\lora_training\\llama.cpp\\common\\download.cpp",281      "preview": "local_path, token, false)) { throw std::runtime_error(\"Failed to download Docker Model\"); } LOG_INF(\"%s: Downloaded Docker Model to: %s\\n\", __func__, local_path.c_str()); return local_path; } catch (const std::exception & e) { LOG_ERR(\"%s: "282    }283  ]284}