Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Baseline the unsloth-zoo 2026.10.7 scan findings after review (#14311)

Daniel Hane326020e4b1f
diff --git a/scripts/scan_packages_baseline.json b/scripts/scan_packages_baseline.json
index 1af7f83f86f..a4eeafc3b10 100644
--- a/scripts/scan_packages_baseline.json
+++ b/scripts/scan_packages_baseline.json
@@ -2295,6 +2295,15 @@
"evidence_hash": "4cfe9a930a03795a0da0693c38733ec6c7f998cdb4e850f2b5a5e8bb469b70af",
"file_sha256": "e442620ac16bd57045df944e7d815db25e4a012ca62784aa46d7587c81f8c2c1"
},
+ {
+ "package": "unsloth-zoo",
+ "file": "unsloth_zoo/triton_kernels_compat.py",
+ "check": "Downloads and executes remote code",
+ "severity": "CRITICAL",
+ "evidence": "L147: topk = importlib.import_module(self.name + \".topk\").topk | L157: ragged = importlib.import_module(self.name + \".tensor_details.ragged_tensor\")",
+ "evidence_hash": "9cf0ce1cb8ab319d7fa6e14425736fd8a7734da12536a758a38d4554a31b7721",
+ "file_sha256": "bbe6078489f9f63a08210c024b03edd4c739c25c059712ef4e72007e1b13b887"
+ },
{
"package": "unsloth-zoo",
"file": "unsloth_zoo/vision_utils.py",
@@ -2474,6 +2483,15 @@
"evidence_hash": "88c38db8f2b8b9d6512709dd7a6d7a9a7bf8a3c48007124ff3680da3a5331d1d",
"file_sha256": "dae0d987ab8566caf0abec70cc3d98cf75509fe7f917a3b3ee04fe5dbc32873c"
},
+ {
+ "package": "unsloth-zoo",
+ "file": "unsloth_zoo/llama_cpp.py",
+ "check": "Harvests environment variables/secrets AND makes network calls",
+ "severity": "CRITICAL",
+ "evidence": "Env: L1040: token = os.environ.get(\"GH_TOKEN\") or os.environ.get(\"GITHUB_TOKEN\") | L5679: env = dict(env) if env is not None else dict(os.environ) | L5766: env = dict(os.environ)\nNetwork: L1048: response = requests.get(url, timeout = timeout, headers = headers, stream = stream) | L4378: response = requests.get(\nL4379: LLAMA_CPP_CONVERT_FILE, timeout = (10, 120)\nL4380: ) | L7843: check = requests.get(llama_cpp_chat_file, timeout = 5)",
+ "evidence_hash": "88c38db8f2b8b9d6512709dd7a6d7a9a7bf8a3c48007124ff3680da3a5331d1d",
+ "file_sha256": "f235013ae6524c7d6d38d78f34e745eef4dea077edfbbed0e2f94bb8b5295cac"
+ },
{
"package": "unsloth-zoo",
"file": "unsloth_zoo/vision_utils.py",
@@ -2643,6 +2661,15 @@
"evidence_hash": "aaba0e6f7c289e2d6ee4ac8ff927d691a450fb3913a4245aa0fab585dd28d871",
"file_sha256": "0e87d9e68dfe0336ca28b48bb17d0dc063f9f438b9ca2cbc3027a9e34ee6f543"
},
+ {
+ "package": "unsloth-zoo",
+ "file": "unsloth_zoo/compiler.py",
+ "check": "Advanced obfuscation (marshal/compile/zlib) + exec/eval",
+ "severity": "HIGH",
+ "evidence": "Obfusc: L1918: exec(compile(source, file_location, \"exec\"), new_module.__dict__) | L2181: _mod = __import__(model_location, fromlist=items) | L7021: f\" {chr(92)}{chr(92)} /| Num examples = {num_examples:,} | Num Epochs = {num_train_epochs:,} | Total steps = {max_steps:,}\\\\n\"\\\\\nL7022: f\"O^O/ {chr(92)}_/ {chr(92)} Batch size per device = {self._train_batch_size:,} | Gradient accumulation steps = {args.gradient_accumulation_steps}\\\\n\"\\\\\nL7023: f\"{chr(92)} / Data Parallel GPUs = {args.world_size} | Total batch size ({self._train_batch_size} x {args.gradient_accumulation_steps} x {args.world_size}) = {total_train_batch_size: sha256:b7ea9cdbe360ad323911014caf12a9d4ec87cb36195a511080e4e484c2fb80d2\nL7024: f' \"-____-\" Trainable parameters = {get_model_param_count(model, trainable_only=True):,} of {get_model_param_count(model):,} ({get_model_param_count(model, trainable_only=True)/get_model_p sha256:9e832c1e7b2815aa44dc42638d821cb9df22550db88d85bd4bfb29c13f984b53 | L7022: f\"O^O/ {chr(92)}_/ {chr(92)} Batch size per device = {self._train_batch_size:,} | Gradient accumulation steps = {args.gradient_accumulation_steps}\\\\n\"\\\\\nL7023: f\"{chr(92)} / Data Parallel GPUs = {args.world_size} | Total batch size ({self._train_batch_size} x {args.gradient_accumulation_steps} x {args.world_size}) = {total_train_batch_size: sha256:b7ea9cdbe360ad323911014caf12a9d4ec87cb36195a511080e4e484c2fb80d2\nL7024: f' \"-____-\" Trainable parameters = {get_model_param_count(model, trainable_only=True):,} of {get_model_param_count(model):,} ({get_model_param_count(model, trainable_only=True)/get_model_p sha256:9e832c1e7b2815aa44dc42638d821cb9df22550db88d85bd4bfb29c13f984b53 | L7021: f\" {chr(92)}{chr(92)} /| Num examples = {num_examples:,} | Num Epochs = {num_train_epochs:,} | Total steps = {max_steps:,}\\\\n\"\\\\\nL7022: f\"O^O/ {chr(92)}_/ {chr(92)} Batch size per device = {self._train_batch_size:,} | Gradient accumulation steps = {args.gradient_accumulation_steps}\\\\n\"\\\\\nL7023: f\"{chr(92)} / Data Parallel GPUs = {args.world_size} | Total batch size ({self._train_batch_size} x {args.gradient_accumulation_steps} x {args.world_size}) = {total_train_batch_size: sha256:b7ea9cdbe360ad323911014caf12a9d4ec87cb36195a511080e4e484c2fb80d2\nExec: L1918: exec(compile(source, file_location, \"exec\"), new_module.__dict__) | L2099: _modeling_file = eval(model_location) | L2837: f = eval(f\"{model_location}.{module}\") | L3196: exec(f\"def raise_{j}(*args, **kwargs): print('{function}')\", globals(), locals()) | L3197: try: exec(f\"EMPTY_LOGITS.{function} = raise_{j}\", globals(), locals()) | L4950: exec(f\"import {parent}\", locals(), globals()) | L5056: dir(eval(parent)), | L5062: exec(f\"{parent}.{child}.forward = forward\", globals(), locals()) | L5142: module = eval(f\"modeling_file.{module}\") | L5169: inner_class = eval(f\"modeling_file.{inner_class}\") | L5404: exec(f\"from timm.layers.norm_act import {norm}\") | L5412: forward = eval(norm).forward | L5418: exec(f\"timm.layers.norm_act.{norm}.forward = forward\") | L5435: exec(f\"from timm.models._efficientnet_blocks import {block}\") | L5443: forward = eval(block).forward | L5449: exec(f\"timm.models._efficientnet_blocks.{block}.forward = forward\") | L5693: exec(\nL5694: f\"{model_location}.torch.nn.{module}.forward = forward\",\nL5695: globals(),\nL5696: locals(),\nL5697: ) | L5699: exec(f\"{model_location}.nn.{module}.forward = forward\", globals(), locals()) | L5703: exec(\nL5704: f\"combined_module.torch.nn.{module}.forward = forward\",\nL5705: globals(),\nL5706: locals(),\nL5707: ) | L5709: exec(f\"combined_module.nn.{module}.forward = forward\", globals(), locals()) | L5786: source = eval(f\"{model_location}.torch\") | L5793: function = eval(f\"source.nn.{module}\") | L5987: exec(\"import transformers\", globals()) | L6004: exec(\nL6005: \"model_logger.addFilter(HideLoggingMessage('`use_cache`'))\", globals(), locals()\nL6006: ) | L6008: exec(\nL6009: \"model_logger.addFilter(HideLoggingMessage('compile_config'))\",\nL6010: globals(),\nL6011: locals(),\nL6012: ) | L6246: source = eval(f\"modeling_file.{module}\") | L6260: source = eval(f\"modeling_file.{module}\") | L6373: source = eval(f\"modeling_file.{module}\") | L6411: source = eval(f\"{model_location}.{module}\") | L6491: source = eval(f\"{model_location}.{module}\") | L6539: source = eval(f\"{model_location}.{module}\") | L6777: source = eval(f\"{model_location}.{module}\") | L6788: exec(\nL6789: f\"{model_location}.{module}._update_causal_mask = no_update_causal_mask\",\nL6790: globals(),\nL6791: ) | L6851: source = eval(f\"{model_location}.{module}\") | L6892: module_cls = eval(f\"{model_location}.{module}\") | L6929: module_cls = eval(f\"{model_location}.{module}\") | L7006: exec(\nL7007: \"from transformers.trainer import (\" + \", \".join(x for x in good_items) + \")\",\nL7008: globals(),\nL7009: ) | L7071: exec(inner_training_loop, globals()) | L7080: function = eval(f\"{model_location}.{module}\") | L7172: function = eval(f\"{model_location}.{module}\") | L7356: exec(\nL7357: f\"{model_location}.{module} = combined_module.{module}\",\nL7358: globals(),\nL7359: locals(),\nL7360: ) | L7370: check_dicts = dir(eval(f\"{model_location}\")) | L7372: item = eval(f\"{model_location}.{check}\") | L7383: exec(\nL7384: f\"{model_location}.{check}['{key}'] = combined_module.{replaced_class}\",\nL7385: globals(),\nL7386: locals(),\nL7387: )",
+ "evidence_hash": "aaba0e6f7c289e2d6ee4ac8ff927d691a450fb3913a4245aa0fab585dd28d871",
+ "file_sha256": "086b2bbef8867ea629bf0d6dd494d2ed7ee72c1259c88b2ce11df7f72850bbc0"
+ },
{
"package": "unsloth-zoo",
"file": "unsloth_zoo/vllm_gdn_replayssm.py",
@@ -2661,6 +2688,15 @@
"evidence_hash": "4928e730b84a3fa71dcc50cfe185c77c2028be09ad908706710ab864786b4659",
"file_sha256": "440a5a8726b5e346d54a7ff3019cdf0413bee4267f6a683e6561660271eed81d"
},
+ {
+ "package": "unsloth-zoo",
+ "file": "unsloth_zoo/vllm_utils.py",
+ "check": "Advanced obfuscation (marshal/compile/zlib) + exec/eval",
+ "severity": "HIGH",
+ "evidence": "Obfusc: L5416: lora_types.append(getattr(__import__(module_path, fromlist = [class_name]), class_name))\nExec: L1419: cross_attn_layer = eval(f'vllm_internals.{name}') | L1464: layernorm = eval(vllm_name).state_dict()[\"weight\"] | L2066: parent = eval(f\"new_model.{parent_path}\") if parent_path else new_model | L2072: exec(f\"new_model.{layer_name_br} = layer\") | L2155: exec(f\"new_model.{layer_name_br}.weight = None\") | L2156: exec(f\"new_model.{layer_name_br}.weight = weight_param\") | L2158: exec(f\"new_model.{layer_name_br}.bias = None\") | L2159: exec(f\"new_model.{layer_name_br}.bias = bias\") | L2165: exec(f\"new_model.{layer_name} = layer\") | L2197: new_model.eval() | L5223: exec(f\"module.to({dtype})\") | L5246: exec(f\"module.to({dtype})\") | L7127: model.eval() | L7167: new_model.eval()",
+ "evidence_hash": "4928e730b84a3fa71dcc50cfe185c77c2028be09ad908706710ab864786b4659",
+ "file_sha256": "7113efd94a0c22ce49d2dc99fbbc80a0a4186cda4d7a8301a9c87fe546a14edc"
+ },
{
"package": "unsloth-zoo",
"file": "unsloth_zoo/mlx/decision.py",
@@ -2679,6 +2715,15 @@
"evidence_hash": "df3ad4dc00bd93c1131bcba7020ff6054aea106e07ebf8ff31dd46715e934670",
"file_sha256": "92834740e88ecbc5269ceb2a7969d669d3874e00f9b8d1e383c7a61904a1655d"
},
+ {
+ "package": "unsloth-zoo",
+ "file": "unsloth_zoo/mlx/loader.py",
+ "check": "Advanced obfuscation (marshal/compile/zlib) + exec/eval",
+ "severity": "HIGH",
+ "evidence": "Obfusc: L5471: _mod = __import__(module_name, fromlist=[\"_\"])\nExec: L161: mx.eval(model.parameters()) | L195: mx.eval([v for k, v in tree_flatten(casted) if is_mlx_norm_parameter_path(k)]) | L1498: mx.eval(model.parameters()) | L1499: mx.eval(mx.distributed.all_sum(mx.array(1.0), stream=mx.cpu)) | L1566: mx.eval(model.parameters()) | L1569: mx.eval(mx.distributed.all_sum(mx.array(1.0), stream=mx.cpu)) | L2998: mx.eval(altup.correct(predictions, activated)) | L3470: model.eval() | L3547: mx.eval(baseline, perturbed) | L3613: mx.eval([value for table in tables for value in table.values()]) | L5119: model.eval() | L5898: mx.eval(model.parameters()) | L6014: mx.eval(module.weight) | L8657: mx.eval(model.parameters()) | L8674: mx.eval(target.parameters()) | L9678: lambda: mx.eval(model.parameters()), | L9742: lambda: mx.eval(model.parameters()), | L9905: mx.eval(model.parameters())",
+ "evidence_hash": "df3ad4dc00bd93c1131bcba7020ff6054aea106e07ebf8ff31dd46715e934670",
+ "file_sha256": "5f9667c4950dd2a25455a98d5fb22a580512a6648b1d000c9d22ce758f321ed7"
+ },
{
"package": "unsloth-zoo",
"file": "unsloth_zoo/mlx/speculative.py",
@@ -2688,6 +2733,15 @@
"evidence_hash": "64b95579c569b39eddc493ac6b808258094d9b1d62154c55ae9c224c80577e45",
"file_sha256": "805e6992d7c03e7c694866904ee944cbd1a7d500d310e1b8d92ea1f3d1fcf1ba"
},
+ {
+ "package": "unsloth-zoo",
+ "file": "unsloth_zoo/mlx/speculative.py",
+ "check": "Advanced obfuscation (marshal/compile/zlib) + exec/eval",
+ "severity": "HIGH",
+ "evidence": "Obfusc: L837: module = __import__(f\"mlx_vlm.models.{family}.language\", fromlist = [\"LanguageModel\"])\nExec: L1084: mx.eval(drafts) | L1108: mx.eval([array for pair in drawn for array in pair if array is not None] + ([] if drafts is None else [drafts])) | L1383: mx.eval(parts[-1]) | L1385: mx.eval(self._scan) | L1878: mx.eval(model.parameters())",
+ "evidence_hash": "64b95579c569b39eddc493ac6b808258094d9b1d62154c55ae9c224c80577e45",
+ "file_sha256": "85fc3f63a2f867a2ab709b8554b7a5fada4c5490f1ce8a88023159c6b4abc761"
+ },
{
"package": "unsloth-zoo",
"file": "unsloth_zoo/temporary_patches/moe_utils.py",
@@ -2696,6 +2750,15 @@
"evidence": "Obfusc: L338: exec(compile(cached_bytes, cache_file, \"exec\"), module.__dict__)\nExec: L338: exec(compile(cached_bytes, cache_file, \"exec\"), module.__dict__)",
"evidence_hash": "ef22a4fe0d8ec7e2b0e4afc6e6a3d997cd94c981bf59790562217d4f9dd70732",
"file_sha256": "cc0ddeab46ef13e61470ea4929b6dba99e7d64cd2269d0f416346006146ffe2c"
+ },
+ {
+ "package": "unsloth-zoo",
+ "file": "unsloth_zoo/temporary_patches/moe_utils.py",
+ "check": "Advanced obfuscation (marshal/compile/zlib) + exec/eval",
+ "severity": "HIGH",
+ "evidence": "Obfusc: L338: exec(compile(cached_bytes, cache_file, \"exec\"), module.__dict__)\nExec: L338: exec(compile(cached_bytes, cache_file, \"exec\"), module.__dict__)",
+ "evidence_hash": "ef22a4fe0d8ec7e2b0e4afc6e6a3d997cd94c981bf59790562217d4f9dd70732",
+ "file_sha256": "00166745336b3f25b244843463445171b86833655cb69593b31e5b7779115416"
}
]
}