hiyouga · yafshar · Oct 24, 2024 · Oct 25, 2024 · Nov 5, 2024 · Nov 16, 2024
diff --git a/examples/inference/llama3.yaml b/examples/inference/llama3.yaml
@@ -1,2 +1,3 @@
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 template: llama3
+trust_remote_code: true
diff --git a/examples/inference/llama3_lora_sft.yaml b/examples/inference/llama3_lora_sft.yaml
@@ -2,3 +2,4 @@ model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
 template: llama3
 finetuning_type: lora
+trust_remote_code: true
diff --git a/examples/inference/llama3_vllm.yaml b/examples/inference/llama3_vllm.yaml
@@ -2,3 +2,4 @@ model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 template: llama3
 infer_backend: vllm
 vllm_enforce_eager: true
+trust_remote_code: true
diff --git a/examples/inference/llava1_5.yaml b/examples/inference/llava1_5.yaml
@@ -1,2 +1,3 @@
 model_name_or_path: llava-hf/llava-1.5-7b-hf
 template: llava
+trust_remote_code: true
diff --git a/examples/inference/qwen2_vl.yaml b/examples/inference/qwen2_vl.yaml
@@ -1,2 +1,3 @@
 model_name_or_path: Qwen/Qwen2-VL-7B-Instruct
 template: qwen2_vl
+trust_remote_code: true
diff --git a/examples/merge_lora/llama3_gptq.yaml b/examples/merge_lora/llama3_gptq.yaml
@@ -1,6 +1,7 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 template: llama3
+trust_remote_code: true
 
 ### export
 export_dir: models/llama3_gptq

diff --git a/examples/merge_lora/llama3_lora_sft.yaml b/examples/merge_lora/llama3_lora_sft.yaml
@@ -5,6 +5,7 @@ model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
 template: llama3
 finetuning_type: lora
+trust_remote_code: true
 
 ### export
 export_dir: models/llama3_lora_sft

diff --git a/examples/merge_lora/qwen2vl_lora_sft.yaml b/examples/merge_lora/qwen2vl_lora_sft.yaml
@@ -5,6 +5,7 @@ model_name_or_path: Qwen/Qwen2-VL-7B-Instruct
 adapter_name_or_path: saves/qwen2_vl-7b/lora/sft
 template: qwen2_vl
 finetuning_type: lora
+trust_remote_code: true
 
 ### export
 export_dir: models/qwen2_vl_lora_sft

diff --git a/examples/train_full/llama3_full_predict.yaml b/examples/train_full/llama3_full_predict.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: saves/llama3-8b/full/sft
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_full/llama3_full_sft_ds3.yaml b/examples/train_full/llama3_full_sft_ds3.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_full/qwen2vl_full_sft.yaml b/examples/train_full/qwen2vl_full_sft.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: Qwen/Qwen2-VL-7B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_lora/llama3_lora_dpo.yaml b/examples/train_lora/llama3_lora_dpo.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: dpo

diff --git a/examples/train_lora/llama3_lora_eval.yaml b/examples/train_lora/llama3_lora_eval.yaml
@@ -1,6 +1,7 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
+trust_remote_code: true
 
 ### method
 finetuning_type: lora

diff --git a/examples/train_lora/llama3_lora_kto.yaml b/examples/train_lora/llama3_lora_kto.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: kto

diff --git a/examples/train_lora/llama3_lora_ppo.yaml b/examples/train_lora/llama3_lora_ppo.yaml
@@ -1,6 +1,7 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 reward_model: saves/llama3-8b/lora/reward
+trust_remote_code: true
 
 ### method
 stage: ppo

diff --git a/examples/train_lora/llama3_lora_predict.yaml b/examples/train_lora/llama3_lora_predict.yaml
@@ -1,6 +1,7 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 adapter_name_or_path: saves/llama3-8b/lora/sft
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_lora/llama3_lora_pretrain.yaml b/examples/train_lora/llama3_lora_pretrain.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: pt

diff --git a/examples/train_lora/llama3_lora_reward.yaml b/examples/train_lora/llama3_lora_reward.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: rm

diff --git a/examples/train_lora/llama3_lora_sft.yaml b/examples/train_lora/llama3_lora_sft.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_lora/llama3_lora_sft_ds0.yaml b/examples/train_lora/llama3_lora_sft_ds0.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_lora/llama3_lora_sft_ds3.yaml b/examples/train_lora/llama3_lora_sft_ds3.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_lora/llama3_preprocess.yaml b/examples/train_lora/llama3_preprocess.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_lora/llava1_5_lora_sft.yaml b/examples/train_lora/llava1_5_lora_sft.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: llava-hf/llava-1.5-7b-hf
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_lora/qwen2vl_lora_dpo.yaml b/examples/train_lora/qwen2vl_lora_dpo.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: Qwen/Qwen2-VL-7B-Instruct
+trust_remote_code: true
 
 ### method
 stage: dpo

diff --git a/examples/train_lora/qwen2vl_lora_sft.yaml b/examples/train_lora/qwen2vl_lora_sft.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: Qwen/Qwen2-VL-7B-Instruct
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_qlora/llama3_lora_sft_aqlm.yaml b/examples/train_qlora/llama3_lora_sft_aqlm.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: ISTA-DASLab/Meta-Llama-3-8B-Instruct-AQLM-2Bit-1x16
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_qlora/llama3_lora_sft_awq.yaml b/examples/train_qlora/llama3_lora_sft_awq.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: TechxGenus/Meta-Llama-3-8B-Instruct-AWQ
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_qlora/llama3_lora_sft_gptq.yaml b/examples/train_qlora/llama3_lora_sft_gptq.yaml
@@ -1,5 +1,6 @@
 ### model
 model_name_or_path: TechxGenus/Meta-Llama-3-8B-Instruct-GPTQ
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/examples/train_qlora/llama3_lora_sft_otfq.yaml b/examples/train_qlora/llama3_lora_sft_otfq.yaml
@@ -2,6 +2,7 @@
 model_name_or_path: meta-llama/Meta-Llama-3-8B-Instruct
 quantization_bit: 4
 quantization_method: bitsandbytes  # choices: [bitsandbytes (4/8), hqq (2/3/4/5/6/8), eetq (8)]
+trust_remote_code: true
 
 ### method
 stage: sft

diff --git a/src/llamafactory/chat/vllm_engine.py b/src/llamafactory/chat/vllm_engine.py
@@ -71,7 +71,7 @@ def __init__(
 
         engine_args = {
             "model": model_args.model_name_or_path,
-            "trust_remote_code": True,
+            "trust_remote_code": model_args.trust_remote_code,
             "download_dir": model_args.cache_dir,
             "dtype": model_args.infer_dtype,
             "max_model_len": model_args.vllm_maxlen,

diff --git a/src/llamafactory/data/loader.py b/src/llamafactory/data/loader.py
@@ -128,7 +128,7 @@ def _load_single_dataset(
             cache_dir=model_args.cache_dir,
             token=model_args.hf_hub_token,
             streaming=data_args.streaming,
-            trust_remote_code=True,
+            trust_remote_code=model_args.trust_remote_code,
         )
 
     if dataset_attr.num_samples is not None and not data_args.streaming:

diff --git a/src/llamafactory/eval/evaluator.py b/src/llamafactory/eval/evaluator.py
@@ -100,7 +100,7 @@ def eval(self) -> None:
                 cache_dir=self.model_args.cache_dir,
                 download_mode=self.eval_args.download_mode,
                 token=self.model_args.hf_hub_token,
-                trust_remote_code=True,
+                trust_remote_code=self.model_args.trust_remote_code,
             )
             pbar.set_postfix_str(categorys[subject]["name"])
             inputs, outputs, labels = [], [], []

diff --git a/src/llamafactory/hparams/model_args.py b/src/llamafactory/hparams/model_args.py
@@ -301,6 +301,16 @@ class ModelArguments(QuantizationArguments, ProcessorArguments, ExportArguments,
         init=False,
         metadata={"help": "Whether use block diag attention or not, derived from `neat_packing`. Do not specify it."},
     )
+    trust_remote_code: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "Whether to trust the execution of code from datasets/models defined on the Hub. "
+                "This option should only be set to `True` for repositories you trust and in which "
+                "you have read the code, as it will execute code present on the Hub on your local machine."
+            )
+        },
+    )
 
     def __post_init__(self):
         if self.model_name_or_path is None:

diff --git a/src/llamafactory/model/loader.py b/src/llamafactory/model/loader.py
@@ -52,7 +52,7 @@ def _get_init_kwargs(model_args: "ModelArguments") -> Dict[str, Any]:
     skip_check_imports()
     model_args.model_name_or_path = try_download_model_from_other_hub(model_args)
     return {
-        "trust_remote_code": True,
+        "trust_remote_code": model_args.trust_remote_code,
         "cache_dir": model_args.cache_dir,
         "revision": model_args.model_revision,
         "token": model_args.hf_hub_token,
@@ -155,8 +155,9 @@ def load_model(
                 load_class = AutoModelForCausalLM
 
             if model_args.train_from_scratch:
-                model = load_class.from_config(config, trust_remote_code=True)
+                model = load_class.from_config(config, trust_remote_code=model_args.trust_remote_code)
             else:
+                init_kwargs["trust_remote_code"] = model_args.trust_remote_code
                 model = load_class.from_pretrained(**init_kwargs)
 
         if model_args.mixture_of_depths == "convert":

diff --git a/src/llamafactory/model/model_utils/unsloth.py b/src/llamafactory/model/model_utils/unsloth.py
@@ -39,7 +39,7 @@ def _get_unsloth_kwargs(
         "device_map": {"": get_current_device()},
         "rope_scaling": getattr(config, "rope_scaling", None),
         "fix_tokenizer": False,
-        "trust_remote_code": True,
+        "trust_remote_code": model_args.trust_remote_code,
         "use_gradient_checkpointing": "unsloth",
     }