v1.0

scripts/llamafy_qwen.py

+# coding=utf-8
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+from collections import OrderedDict
+from typing import Any, Dict
+
+import fire
+import torch
+from safetensors import safe_open
+from safetensors.torch import save_file
+from tqdm import tqdm
+from transformers.modeling_utils import (
+    SAFE_WEIGHTS_INDEX_NAME,
+    SAFE_WEIGHTS_NAME,
+    WEIGHTS_INDEX_NAME,
+    WEIGHTS_NAME,
+    shard_checkpoint,
+)
+from transformers.utils import check_min_version
+
+
+try:
+    check_min_version("4.34.0")
+except Exception:
+    raise ValueError("Please upgrade `transformers` to 4.34.0")
+
+
+CONFIG_NAME = "config.json"
+
+
+def save_weight(input_dir: str, output_dir: str, shard_size: str, save_safetensors: bool) -> str:
+    qwen_state_dict: Dict[str, torch.Tensor] = OrderedDict()
+    for filepath in tqdm(os.listdir(input_dir), desc="Load weights"):
+        if os.path.isfile(os.path.join(input_dir, filepath)) and filepath.endswith(".safetensors"):
+            with safe_open(os.path.join(input_dir, filepath), framework="pt", device="cpu") as f:
+                for key in f.keys():
+                    qwen_state_dict[key] = f.get_tensor(key)
+
+    llama2_state_dict: Dict[str, torch.Tensor] = OrderedDict()
+    torch_dtype = None
+    for key, value in tqdm(qwen_state_dict.items(), desc="Convert format"):
+        if torch_dtype is None:
+            torch_dtype = value.dtype
+        if "wte" in key:
+            llama2_state_dict["model.embed_tokens.weight"] = value
+        elif "ln_f" in key:
+            llama2_state_dict["model.norm.weight"] = value
+        else:
+            key = key.replace("transformer.h", "model.layers")
+            if "attn.c_attn" in key:
+                proj_size = value.size(0) // 3
+                llama2_state_dict[key.replace("attn.c_attn", "self_attn.q_proj")] = value[:proj_size, ...]
+                llama2_state_dict[key.replace("attn.c_attn", "self_attn.k_proj")] = value[
+                    proj_size : 2 * proj_size, ...
+                ]
+                llama2_state_dict[key.replace("attn.c_attn", "self_attn.v_proj")] = value[2 * proj_size :, ...]
+            elif "attn.c_proj" in key:
+                llama2_state_dict[key.replace("attn.c_proj", "self_attn.o_proj")] = value
+                llama2_state_dict[key.replace("attn.c_proj.weight", "self_attn.o_proj.bias")] = torch.zeros_like(
+                    value[:, 0]
+                ).squeeze()
+            elif "ln_1" in key:
+                llama2_state_dict[key.replace("ln_1", "input_layernorm")] = value
+            elif "ln_2" in key:
+                llama2_state_dict[key.replace("ln_2", "post_attention_layernorm")] = value
+            elif "mlp.w1" in key:
+                llama2_state_dict[key.replace("mlp.w1", "mlp.up_proj")] = value
+            elif "mlp.w2" in key:
+                llama2_state_dict[key.replace("mlp.w2", "mlp.gate_proj")] = value
+            elif "mlp.c_proj" in key:
+                llama2_state_dict[key.replace("mlp.c_proj", "mlp.down_proj")] = value
+            elif "lm_head" in key:
+                llama2_state_dict[key] = value
+            else:
+                raise KeyError("Unable to process key {}".format(key))
+
+    weights_name = SAFE_WEIGHTS_NAME if save_safetensors else WEIGHTS_NAME
+    shards, index = shard_checkpoint(llama2_state_dict, max_shard_size=shard_size, weights_name=weights_name)
+
+    for shard_file, shard in tqdm(shards.items(), desc="Save weights"):
+        if save_safetensors:
+            save_file(shard, os.path.join(output_dir, shard_file), metadata={"format": "pt"})
+        else:
+            torch.save(shard, os.path.join(output_dir, shard_file))
+
+    if index is None:
+        print("Model weights saved in {}".format(os.path.join(output_dir, weights_name)))
+    else:
+        index_name = SAFE_WEIGHTS_INDEX_NAME if save_safetensors else WEIGHTS_INDEX_NAME
+        with open(os.path.join(output_dir, index_name), "w", encoding="utf-8") as f:
+            json.dump(index, f, indent=2, sort_keys=True)
+        print("Model weights saved in {}".format(output_dir))
+
+    return str(torch_dtype).replace("torch.", "")
+
+
+def save_config(input_dir: str, output_dir: str, torch_dtype: str):
+    with open(os.path.join(input_dir, CONFIG_NAME), "r", encoding="utf-8") as f:
+        qwen_config_dict: Dict[str, Any] = json.load(f)
+
+    llama2_config_dict: Dict[str, Any] = OrderedDict()
+    llama2_config_dict["architectures"] = ["LlamaForCausalLM"]
+    llama2_config_dict["hidden_act"] = "silu"
+    llama2_config_dict["hidden_size"] = qwen_config_dict["hidden_size"]
+    llama2_config_dict["initializer_range"] = qwen_config_dict["initializer_range"]
+    llama2_config_dict["intermediate_size"] = qwen_config_dict["intermediate_size"] // 2
+    llama2_config_dict["max_position_embeddings"] = qwen_config_dict["max_position_embeddings"]
+    llama2_config_dict["model_type"] = "llama"
+    llama2_config_dict["num_attention_heads"] = qwen_config_dict["num_attention_heads"]
+    llama2_config_dict["num_hidden_layers"] = qwen_config_dict["num_hidden_layers"]
+    llama2_config_dict["num_key_value_heads"] = qwen_config_dict["hidden_size"] // qwen_config_dict["kv_channels"]
+    llama2_config_dict["pretraining_tp"] = 1
+    llama2_config_dict["rms_norm_eps"] = qwen_config_dict["layer_norm_epsilon"]
+    llama2_config_dict["rope_scaling"] = None
+    llama2_config_dict["tie_word_embeddings"] = qwen_config_dict["tie_word_embeddings"]
+    llama2_config_dict["torch_dtype"] = torch_dtype
+    llama2_config_dict["transformers_version"] = "4.34.0"
+    llama2_config_dict["use_cache"] = True
+    llama2_config_dict["vocab_size"] = qwen_config_dict["vocab_size"]
+    llama2_config_dict["attention_bias"] = True
+
+    with open(os.path.join(output_dir, CONFIG_NAME), "w", encoding="utf-8") as f:
+        json.dump(llama2_config_dict, f, indent=2)
+    print("Model config saved in {}".format(os.path.join(output_dir, CONFIG_NAME)))
+
+
+def llamafy_qwen(
+    input_dir: str,
+    output_dir: str,
+    shard_size: str = "2GB",
+    save_safetensors: bool = False,
+):
+    r"""
+    Converts the Qwen models in the same format as LLaMA2.
+    Usage: python llamafy_qwen.py --input_dir input --output_dir output
+    Converted model: https://huggingface.co/hiyouga/Qwen-14B-Chat-LLaMAfied
+    """
+    try:
+        os.makedirs(output_dir, exist_ok=False)
+    except Exception as e:
+        raise print("Output dir already exists", e)
+
+    torch_dtype = save_weight(input_dir, output_dir, shard_size, save_safetensors)
+    save_config(input_dir, output_dir, torch_dtype)
+
+
+if __name__ == "__main__":
+    fire.Fire(llamafy_qwen)

scripts/loftq_init.py

+# coding=utf-8
+# Copyright 2024 HuggingFace Inc. and the LlamaFactory team.
+#
+# This code is based on the HuggingFace's PEFT library.
+# https://github.com/huggingface/peft/blob/v0.10.0/examples/loftq_finetuning/quantize_save_load.py
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from typing import TYPE_CHECKING
+
+import fire
+from peft import LoftQConfig, LoraConfig, TaskType, get_peft_model
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedModel
+
+
+def quantize_loftq(
+    model_name_or_path: str,
+    output_dir: str,
+    loftq_bits: int = 4,
+    loftq_iter: int = 4,
+    lora_alpha: int = None,
+    lora_rank: int = 16,
+    lora_dropout: float = 0,
+    lora_target: tuple = ("q_proj", "v_proj"),
+    save_safetensors: bool = True,
+):
+    r"""
+    Initializes LoRA weights with LoRA-fine-tuning-aware Quantization (LoftQ)
+    Usage: python loftq_init.py --model_name_or_path path_to_model --output_dir output_dir
+    """
+    if isinstance(lora_target, str):
+        lora_target = [name.strip() for name in lora_target.split(",")]
+
+    tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=True)
+    model = AutoModelForCausalLM.from_pretrained(model_name_or_path, trust_remote_code=True, torch_dtype="auto")
+
+    loftq_config = LoftQConfig(loftq_bits=loftq_bits, loftq_iter=loftq_iter)
+    lora_config = LoraConfig(
+        task_type=TaskType.CAUSAL_LM,
+        inference_mode=True,
+        r=lora_rank,
+        lora_alpha=lora_alpha if lora_alpha is not None else lora_rank * 2,
+        lora_dropout=lora_dropout,
+        target_modules=lora_target,
+        init_lora_weights="loftq",
+        loftq_config=loftq_config,
+    )
+
+    # Init LoftQ model
+    print("Initializing LoftQ weights, it may be take several minutes, wait patiently.")
+    peft_model = get_peft_model(model, lora_config)
+    loftq_dir = os.path.join(output_dir, "loftq_init")
+
+    # Save LoftQ model
+    setattr(peft_model.peft_config["default"], "base_model_name_or_path", os.path.abspath(output_dir))
+    setattr(peft_model.peft_config["default"], "init_lora_weights", True)  # don't apply loftq again
+    peft_model.save_pretrained(loftq_dir, safe_serialization=save_safetensors)
+    print("Adapter weights saved in {}".format(loftq_dir))
+
+    # Save base model
+    base_model: "PreTrainedModel" = peft_model.unload()
+    base_model.save_pretrained(output_dir, safe_serialization=save_safetensors)
+    tokenizer.save_pretrained(output_dir)
+    print("Model weights saved in {}".format(output_dir))
+
+    print("- Fine-tune this model with:")
+    print("model_name_or_path: {}".format(output_dir))
+    print("adapter_name_or_path: {}".format(loftq_dir))
+    print("finetuning_type: lora")
+    print("quantization_bit: {}".format(loftq_bits))
+
+
+if __name__ == "__main__":
+    fire.Fire(quantize_loftq)

scripts/pissa_init.py

+# coding=utf-8
+# Copyright 2024 HuggingFace Inc. and the LlamaFactory team.
+#
+# This code is based on the HuggingFace's PEFT library.
+# https://github.com/huggingface/peft/blob/v0.11.0/examples/pissa_finetuning/preprocess.py
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from typing import TYPE_CHECKING
+
+import fire
+from peft import LoraConfig, TaskType, get_peft_model
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedModel
+
+
+def quantize_pissa(
+    model_name_or_path: str,
+    output_dir: str,
+    pissa_iter: int = 16,
+    lora_alpha: int = None,
+    lora_rank: int = 16,
+    lora_dropout: float = 0,
+    lora_target: tuple = ("q_proj", "v_proj"),
+    save_safetensors: bool = True,
+):
+    r"""
+    Initializes LoRA weights with Principal Singular values and Singular vectors Adaptation (PiSSA)
+    Usage: python pissa_init.py --model_name_or_path path_to_model --output_dir output_dir
+    """
+    if isinstance(lora_target, str):
+        lora_target = [name.strip() for name in lora_target.split(",")]
+
+    tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, trust_remote_code=True)
+    model = AutoModelForCausalLM.from_pretrained(model_name_or_path, trust_remote_code=True, torch_dtype="auto")
+
+    lora_config = LoraConfig(
+        task_type=TaskType.CAUSAL_LM,
+        r=lora_rank,
+        lora_alpha=lora_alpha if lora_alpha is not None else lora_rank * 2,
+        lora_dropout=lora_dropout,
+        target_modules=lora_target,
+        init_lora_weights="pissa" if pissa_iter == -1 else "pissa_niter_{}".format(pissa_iter),
+    )
+
+    # Init PiSSA model
+    peft_model = get_peft_model(model, lora_config)
+    pissa_dir = os.path.join(output_dir, "pissa_init")
+
+    # Save PiSSA model
+    setattr(peft_model.peft_config["default"], "base_model_name_or_path", os.path.abspath(output_dir))
+    setattr(peft_model.peft_config["default"], "init_lora_weights", True)  # don't apply pissa again
+    peft_model.save_pretrained(pissa_dir, safe_serialization=save_safetensors)
+    print("Adapter weights saved in {}".format(pissa_dir))
+
+    # Save base model
+    base_model: "PreTrainedModel" = peft_model.unload()
+    base_model.save_pretrained(output_dir, safe_serialization=save_safetensors)
+    tokenizer.save_pretrained(output_dir)
+    print("Model weights saved in {}".format(output_dir))
+
+    print("- Fine-tune this model with:")
+    print("model_name_or_path: {}".format(output_dir))
+    print("adapter_name_or_path: {}".format(pissa_dir))
+    print("finetuning_type: lora")
+    print("pissa_init: false")
+    print("pissa_convert: true")
+    print("- and optionally with:")
+    print("quantization_bit: 4")
+
+
+if __name__ == "__main__":
+    fire.Fire(quantize_pissa)

scripts/test_toolcall.py

+# coding=utf-8
+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+import os
+from typing import Sequence
+
+from openai import OpenAI
+from transformers.utils.versions import require_version
+
+
+require_version("openai>=1.5.0", "To fix: pip install openai>=1.5.0")
+
+
+def calculate_gpa(grades: Sequence[str], hours: Sequence[int]) -> float:
+    grade_to_score = {"A": 4, "B": 3, "C": 2}
+    total_score, total_hour = 0, 0
+    for grade, hour in zip(grades, hours):
+        total_score += grade_to_score[grade] * hour
+        total_hour += hour
+    return round(total_score / total_hour, 2)
+
+
+def main():
+    client = OpenAI(
+        api_key="{}".format(os.environ.get("API_KEY", "0")),
+        base_url="http://localhost:{}/v1".format(os.environ.get("API_PORT", 8000)),
+    )
+    tools = [
+        {
+            "type": "function",
+            "function": {
+                "name": "calculate_gpa",
+                "description": "Calculate the Grade Point Average (GPA) based on grades and credit hours",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "grades": {"type": "array", "items": {"type": "string"}, "description": "The grades"},
+                        "hours": {"type": "array", "items": {"type": "integer"}, "description": "The credit hours"},
+                    },
+                    "required": ["grades", "hours"],
+                },
+            },
+        }
+    ]
+    tool_map = {"calculate_gpa": calculate_gpa}
+
+    messages = []
+    messages.append({"role": "user", "content": "My grades are A, A, B, and C. The credit hours are 3, 4, 3, and 2."})
+    result = client.chat.completions.create(messages=messages, model="test", tools=tools)
+    if result.choices[0].message.tool_calls is None:
+        raise ValueError("Cannot retrieve function call from the response.")
+
+    messages.append(result.choices[0].message)
+    tool_call = result.choices[0].message.tool_calls[0].function
+    print(tool_call)
+    # Function(arguments='{"grades": ["A", "A", "B", "C"], "hours": [3, 4, 3, 2]}', name='calculate_gpa')
+    name, arguments = tool_call.name, json.loads(tool_call.arguments)
+    tool_result = tool_map[name](**arguments)
+    messages.append({"role": "tool", "content": json.dumps({"gpa": tool_result}, ensure_ascii=False)})
+    result = client.chat.completions.create(messages=messages, model="test", tools=tools)
+    print(result.choices[0].message.content)
+    # Based on the grades and credit hours you provided, your Grade Point Average (GPA) is 3.42.
+
+
+if __name__ == "__main__":
+    main()

setup.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import re
+from typing import List
+
+from setuptools import find_packages, setup
+
+
+def get_version() -> str:
+    with open(os.path.join("src", "llamafactory", "extras", "env.py"), "r", encoding="utf-8") as f:
+        file_content = f.read()
+        pattern = r"{}\W*=\W*\"([^\"]+)\"".format("VERSION")
+        (version,) = re.findall(pattern, file_content)
+        return version
+
+
+def get_requires() -> List[str]:
+    with open("requirements.txt", "r", encoding="utf-8") as f:
+        file_content = f.read()
+        lines = [line.strip() for line in file_content.strip().split("\n") if not line.startswith("#")]
+        return lines
+
+
+def get_console_scripts() -> List[str]:
+    console_scripts = ["llamafactory-cli = llamafactory.cli:main"]
+    if os.environ.get("ENABLE_SHORT_CONSOLE", "1").lower() in ["true", "1"]:
+        console_scripts.append("lmf = llamafactory.cli:main")
+
+    return console_scripts
+
+
+extra_require = {
+    "torch": ["torch>=1.13.1"],
+    "torch-npu": ["torch==2.1.0", "torch-npu==2.1.0.post3", "decorator"],
+    "metrics": ["nltk", "jieba", "rouge-chinese"],
+    "deepspeed": ["deepspeed>=0.10.0,<=0.14.4"],
+    "liger-kernel": ["liger-kernel"],
+    "bitsandbytes": ["bitsandbytes>=0.39.0"],
+    "hqq": ["hqq"],
+    "eetq": ["eetq"],
+    "gptq": ["optimum>=1.17.0", "auto-gptq>=0.5.0"],
+    "awq": ["autoawq"],
+    "aqlm": ["aqlm[gpu]>=1.1.0"],
+    "vllm": ["vllm>=0.4.3,<=0.6.3"],
+    "galore": ["galore-torch"],
+    "badam": ["badam>=1.2.1"],
+    "adam-mini": ["adam-mini"],
+    "qwen": ["transformers_stream_generator"],
+    "modelscope": ["modelscope"],
+    "openmind": ["openmind"],
+    "dev": ["ruff", "pytest"],
+}
+
+
+def main():
+    setup(
+        name="llamafactory",
+        version=get_version(),
+        author="hiyouga",
+        author_email="hiyouga" "@" "buaa.edu.cn",
+        description="Easy-to-use LLM fine-tuning framework",
+        long_description=open("README.md", "r", encoding="utf-8").read(),
+        long_description_content_type="text/markdown",
+        keywords=["LLaMA", "BLOOM", "Falcon", "LLM", "ChatGPT", "transformer", "pytorch", "deep learning"],
+        license="Apache 2.0 License",
+        url="https://github.com/hiyouga/LLaMA-Factory",
+        package_dir={"": "src"},
+        packages=find_packages("src"),
+        python_requires=">=3.8.0",
+        install_requires=get_requires(),
+        extras_require=extra_require,
+        entry_points={"console_scripts": get_console_scripts()},
+        classifiers=[
+            "Development Status :: 4 - Beta",
+            "Intended Audience :: Developers",
+            "Intended Audience :: Education",
+            "Intended Audience :: Science/Research",
+            "License :: OSI Approved :: Apache Software License",
+            "Operating System :: OS Independent",
+            "Programming Language :: Python :: 3",
+            "Programming Language :: Python :: 3.8",
+            "Programming Language :: Python :: 3.9",
+            "Programming Language :: Python :: 3.10",
+            "Programming Language :: Python :: 3.11",
+            "Topic :: Scientific/Engineering :: Artificial Intelligence",
+        ],
+    )
+
+
+if __name__ == "__main__":
+    main()

src/api.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+
+import uvicorn
+
+from llamafactory.api.app import create_app
+from llamafactory.chat import ChatModel
+
+
+def main():
+    chat_model = ChatModel()
+    app = create_app(chat_model)
+    api_host = os.environ.get("API_HOST", "0.0.0.0")
+    api_port = int(os.environ.get("API_PORT", "8000"))
+    print("Visit http://localhost:{}/docs for API document.".format(api_port))
+    uvicorn.run(app, host=api_host, port=api_port)
+
+
+if __name__ == "__main__":
+    main()

src/llamafactory/__init__.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+r"""
+Efficient fine-tuning of large language models.
+
+Level:
+  api, webui > chat, eval, train > data, model > hparams > extras
+
+Dependency graph:
+  main:
+    transformers>=4.41.2,<=4.45.2
+    datasets>=2.16.0,<=2.21.0
+    accelerate>=0.30.1,<=0.34.2
+    peft>=0.11.1,<=0.12.0
+    trl>=0.8.6,<=0.9.6
+  attention:
+    transformers>=4.42.4 (gemma+fa2)
+  longlora:
+    transformers>=4.41.2,<=4.45.2
+  packing:
+    transformers>=4.41.2,<=4.45.2
+
+Disable version checking: DISABLE_VERSION_CHECK=1
+Enable VRAM recording: RECORD_VRAM=1
+Force check imports: FORCE_CHECK_IMPORTS=1
+Force using torchrun: FORCE_TORCHRUN=1
+Set logging verbosity: LLAMAFACTORY_VERBOSITY=WARN
+Use modelscope: USE_MODELSCOPE_HUB=1
+Use openmind: USE_OPENMIND_HUB=1
+"""
+
+from .extras.env import VERSION
+
+
+__version__ = VERSION

src/llamafactory/api/app.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+import os
+from contextlib import asynccontextmanager
+from functools import partial
+from typing import Optional
+
+from typing_extensions import Annotated
+
+from ..chat import ChatModel
+from ..extras.misc import torch_gc
+from ..extras.packages import is_fastapi_available, is_starlette_available, is_uvicorn_available
+from .chat import (
+    create_chat_completion_response,
+    create_score_evaluation_response,
+    create_stream_chat_completion_response,
+)
+from .protocol import (
+    ChatCompletionRequest,
+    ChatCompletionResponse,
+    ModelCard,
+    ModelList,
+    ScoreEvaluationRequest,
+    ScoreEvaluationResponse,
+)
+
+
+if is_fastapi_available():
+    from fastapi import Depends, FastAPI, HTTPException, status
+    from fastapi.middleware.cors import CORSMiddleware
+    from fastapi.security.http import HTTPAuthorizationCredentials, HTTPBearer
+
+
+if is_starlette_available():
+    from sse_starlette import EventSourceResponse
+
+
+if is_uvicorn_available():
+    import uvicorn
+
+
+async def sweeper() -> None:
+    while True:
+        torch_gc()
+        await asyncio.sleep(300)
+
+
+@asynccontextmanager
+async def lifespan(app: "FastAPI", chat_model: "ChatModel"):  # collects GPU memory
+    if chat_model.engine_type == "huggingface":
+        asyncio.create_task(sweeper())
+
+    yield
+    torch_gc()
+
+
+def create_app(chat_model: "ChatModel") -> "FastAPI":
+    root_path = os.environ.get("FASTAPI_ROOT_PATH", "")
+    app = FastAPI(lifespan=partial(lifespan, chat_model=chat_model), root_path=root_path)
+    app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],
+        allow_credentials=True,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+    api_key = os.environ.get("API_KEY", None)
+    security = HTTPBearer(auto_error=False)
+
+    async def verify_api_key(auth: Annotated[Optional[HTTPAuthorizationCredentials], Depends(security)]):
+        if api_key and (auth is None or auth.credentials != api_key):
+            raise HTTPException(status_code=status.HTTP_401_UNAUTHORIZED, detail="Invalid API key.")
+
+    @app.get(
+        "/v1/models",
+        response_model=ModelList,
+        status_code=status.HTTP_200_OK,
+        dependencies=[Depends(verify_api_key)],
+    )
+    async def list_models():
+        model_card = ModelCard(id=os.environ.get("API_MODEL_NAME", "gpt-3.5-turbo"))
+        return ModelList(data=[model_card])
+
+    @app.post(
+        "/v1/chat/completions",
+        response_model=ChatCompletionResponse,
+        status_code=status.HTTP_200_OK,
+        dependencies=[Depends(verify_api_key)],
+    )
+    async def create_chat_completion(request: ChatCompletionRequest):
+        if not chat_model.engine.can_generate:
+            raise HTTPException(status_code=status.HTTP_405_METHOD_NOT_ALLOWED, detail="Not allowed")
+
+        if request.stream:
+            generate = create_stream_chat_completion_response(request, chat_model)
+            return EventSourceResponse(generate, media_type="text/event-stream")
+        else:
+            return await create_chat_completion_response(request, chat_model)
+
+    @app.post(
+        "/v1/score/evaluation",
+        response_model=ScoreEvaluationResponse,
+        status_code=status.HTTP_200_OK,
+        dependencies=[Depends(verify_api_key)],
+    )
+    async def create_score_evaluation(request: ScoreEvaluationRequest):
+        if chat_model.engine.can_generate:
+            raise HTTPException(status_code=status.HTTP_405_METHOD_NOT_ALLOWED, detail="Not allowed")
+
+        return await create_score_evaluation_response(request, chat_model)
+
+    return app
+
+
+def run_api() -> None:
+    chat_model = ChatModel()
+    app = create_app(chat_model)
+    api_host = os.environ.get("API_HOST", "0.0.0.0")
+    api_port = int(os.environ.get("API_PORT", "8000"))
+    print("Visit http://localhost:{}/docs for API document.".format(api_port))
+    uvicorn.run(app, host=api_host, port=api_port)

src/llamafactory/api/chat.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import base64
+import io
+import json
+import os
+import re
+import uuid
+from typing import TYPE_CHECKING, AsyncGenerator, Dict, List, Optional, Tuple
+
+from ..data import Role as DataRole
+from ..extras.logging import get_logger
+from ..extras.packages import is_fastapi_available, is_pillow_available, is_requests_available
+from .common import dictify, jsonify
+from .protocol import (
+    ChatCompletionMessage,
+    ChatCompletionResponse,
+    ChatCompletionResponseChoice,
+    ChatCompletionResponseUsage,
+    ChatCompletionStreamResponse,
+    ChatCompletionStreamResponseChoice,
+    Finish,
+    Function,
+    FunctionCall,
+    Role,
+    ScoreEvaluationResponse,
+)
+
+
+if is_fastapi_available():
+    from fastapi import HTTPException, status
+
+
+if is_pillow_available():
+    from PIL import Image
+
+
+if is_requests_available():
+    import requests
+
+
+if TYPE_CHECKING:
+    from ..chat import ChatModel
+    from ..data.mm_plugin import ImageInput
+    from .protocol import ChatCompletionRequest, ScoreEvaluationRequest
+
+
+logger = get_logger(__name__)
+ROLE_MAPPING = {
+    Role.USER: DataRole.USER.value,
+    Role.ASSISTANT: DataRole.ASSISTANT.value,
+    Role.SYSTEM: DataRole.SYSTEM.value,
+    Role.FUNCTION: DataRole.FUNCTION.value,
+    Role.TOOL: DataRole.OBSERVATION.value,
+}
+
+
+def _process_request(
+    request: "ChatCompletionRequest",
+) -> Tuple[List[Dict[str, str]], Optional[str], Optional[str], Optional["ImageInput"]]:
+    logger.info("==== request ====\n{}".format(json.dumps(dictify(request), indent=2, ensure_ascii=False)))
+
+    if len(request.messages) == 0:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid length")
+
+    if request.messages[0].role == Role.SYSTEM:
+        system = request.messages.pop(0).content
+    else:
+        system = None
+
+    if len(request.messages) % 2 == 0:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Only supports u/a/u/a/u...")
+
+    input_messages = []
+    image = None
+    for i, message in enumerate(request.messages):
+        if i % 2 == 0 and message.role not in [Role.USER, Role.TOOL]:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid role")
+        elif i % 2 == 1 and message.role not in [Role.ASSISTANT, Role.FUNCTION]:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid role")
+
+        if message.role == Role.ASSISTANT and isinstance(message.tool_calls, list) and len(message.tool_calls):
+            tool_calls = [
+                {"name": tool_call.function.name, "arguments": tool_call.function.arguments}
+                for tool_call in message.tool_calls
+            ]
+            content = json.dumps(tool_calls, ensure_ascii=False)
+            input_messages.append({"role": ROLE_MAPPING[Role.FUNCTION], "content": content})
+        elif isinstance(message.content, list):
+            for input_item in message.content:
+                if input_item.type == "text":
+                    input_messages.append({"role": ROLE_MAPPING[message.role], "content": input_item.text})
+                else:
+                    image_url = input_item.image_url.url
+                    if re.match(r"^data:image\/(png|jpg|jpeg|gif|bmp);base64,(.+)$", image_url):  # base64 image
+                        image_stream = io.BytesIO(base64.b64decode(image_url.split(",", maxsplit=1)[1]))
+                    elif os.path.isfile(image_url):  # local file
+                        image_stream = open(image_url, "rb")
+                    else:  # web uri
+                        image_stream = requests.get(image_url, stream=True).raw
+
+                    image = Image.open(image_stream).convert("RGB")
+        else:
+            input_messages.append({"role": ROLE_MAPPING[message.role], "content": message.content})
+
+    tool_list = request.tools
+    if isinstance(tool_list, list) and len(tool_list):
+        try:
+            tools = json.dumps([dictify(tool.function) for tool in tool_list], ensure_ascii=False)
+        except json.JSONDecodeError:
+            raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid tools")
+    else:
+        tools = None
+
+    return input_messages, system, tools, image
+
+
+def _create_stream_chat_completion_chunk(
+    completion_id: str,
+    model: str,
+    delta: "ChatCompletionMessage",
+    index: Optional[int] = 0,
+    finish_reason: Optional["Finish"] = None,
+) -> str:
+    choice_data = ChatCompletionStreamResponseChoice(index=index, delta=delta, finish_reason=finish_reason)
+    chunk = ChatCompletionStreamResponse(id=completion_id, model=model, choices=[choice_data])
+    return jsonify(chunk)
+
+
+async def create_chat_completion_response(
+    request: "ChatCompletionRequest", chat_model: "ChatModel"
+) -> "ChatCompletionResponse":
+    completion_id = "chatcmpl-{}".format(uuid.uuid4().hex)
+    input_messages, system, tools, image = _process_request(request)
+    responses = await chat_model.achat(
+        input_messages,
+        system,
+        tools,
+        image,
+        do_sample=request.do_sample,
+        temperature=request.temperature,
+        top_p=request.top_p,
+        max_new_tokens=request.max_tokens,
+        num_return_sequences=request.n,
+        stop=request.stop,
+    )
+
+    prompt_length, response_length = 0, 0
+    choices = []
+    for i, response in enumerate(responses):
+        if tools:
+            result = chat_model.engine.template.extract_tool(response.response_text)
+        else:
+            result = response.response_text
+
+        if isinstance(result, list):
+            tool_calls = []
+            for tool in result:
+                function = Function(name=tool[0], arguments=tool[1])
+                tool_calls.append(FunctionCall(id="call_{}".format(uuid.uuid4().hex), function=function))
+
+            response_message = ChatCompletionMessage(role=Role.ASSISTANT, tool_calls=tool_calls)
+            finish_reason = Finish.TOOL
+        else:
+            response_message = ChatCompletionMessage(role=Role.ASSISTANT, content=result)
+            finish_reason = Finish.STOP if response.finish_reason == "stop" else Finish.LENGTH
+
+        choices.append(ChatCompletionResponseChoice(index=i, message=response_message, finish_reason=finish_reason))
+        prompt_length = response.prompt_length
+        response_length += response.response_length
+
+    usage = ChatCompletionResponseUsage(
+        prompt_tokens=prompt_length,
+        completion_tokens=response_length,
+        total_tokens=prompt_length + response_length,
+    )
+
+    return ChatCompletionResponse(id=completion_id, model=request.model, choices=choices, usage=usage)
+
+
+async def create_stream_chat_completion_response(
+    request: "ChatCompletionRequest", chat_model: "ChatModel"
+) -> AsyncGenerator[str, None]:
+    completion_id = "chatcmpl-{}".format(uuid.uuid4().hex)
+    input_messages, system, tools, image = _process_request(request)
+    if tools:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Cannot stream function calls.")
+
+    if request.n > 1:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Cannot stream multiple responses.")
+
+    yield _create_stream_chat_completion_chunk(
+        completion_id=completion_id, model=request.model, delta=ChatCompletionMessage(role=Role.ASSISTANT, content="")
+    )
+    async for new_token in chat_model.astream_chat(
+        input_messages,
+        system,
+        tools,
+        image,
+        do_sample=request.do_sample,
+        temperature=request.temperature,
+        top_p=request.top_p,
+        max_new_tokens=request.max_tokens,
+        stop=request.stop,
+    ):
+        if len(new_token) != 0:
+            yield _create_stream_chat_completion_chunk(
+                completion_id=completion_id, model=request.model, delta=ChatCompletionMessage(content=new_token)
+            )
+
+    yield _create_stream_chat_completion_chunk(
+        completion_id=completion_id, model=request.model, delta=ChatCompletionMessage(), finish_reason=Finish.STOP
+    )
+    yield "[DONE]"
+
+
+async def create_score_evaluation_response(
+    request: "ScoreEvaluationRequest", chat_model: "ChatModel"
+) -> "ScoreEvaluationResponse":
+    score_id = "scoreval-{}".format(uuid.uuid4().hex)
+    if len(request.messages) == 0:
+        raise HTTPException(status_code=status.HTTP_400_BAD_REQUEST, detail="Invalid request")
+
+    scores = await chat_model.aget_scores(request.messages, max_length=request.max_length)
+    return ScoreEvaluationResponse(id=score_id, model=request.model, scores=scores)

src/llamafactory/api/common.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import json
+from typing import TYPE_CHECKING, Any, Dict
+
+
+if TYPE_CHECKING:
+    from pydantic import BaseModel
+
+
+def dictify(data: "BaseModel") -> Dict[str, Any]:
+    try:  # pydantic v2
+        return data.model_dump(exclude_unset=True)
+    except AttributeError:  # pydantic v1
+        return data.dict(exclude_unset=True)
+
+
+def jsonify(data: "BaseModel") -> str:
+    try:  # pydantic v2
+        return json.dumps(data.model_dump(exclude_unset=True), ensure_ascii=False)
+    except AttributeError:  # pydantic v1
+        return data.json(exclude_unset=True, ensure_ascii=False)

src/llamafactory/api/protocol.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import time
+from enum import Enum, unique
+from typing import Any, Dict, List, Optional, Union
+
+from pydantic import BaseModel, Field
+from typing_extensions import Literal
+
+
+@unique
+class Role(str, Enum):
+    USER = "user"
+    ASSISTANT = "assistant"
+    SYSTEM = "system"
+    FUNCTION = "function"
+    TOOL = "tool"
+
+
+@unique
+class Finish(str, Enum):
+    STOP = "stop"
+    LENGTH = "length"
+    TOOL = "tool_calls"
+
+
+class ModelCard(BaseModel):
+    id: str
+    object: Literal["model"] = "model"
+    created: int = Field(default_factory=lambda: int(time.time()))
+    owned_by: Literal["owner"] = "owner"
+
+
+class ModelList(BaseModel):
+    object: Literal["list"] = "list"
+    data: List[ModelCard] = []
+
+
+class Function(BaseModel):
+    name: str
+    arguments: str
+
+
+class FunctionDefinition(BaseModel):
+    name: str
+    description: str
+    parameters: Dict[str, Any]
+
+
+class FunctionAvailable(BaseModel):
+    type: Literal["function", "code_interpreter"] = "function"
+    function: Optional[FunctionDefinition] = None
+
+
+class FunctionCall(BaseModel):
+    id: str
+    type: Literal["function"] = "function"
+    function: Function
+
+
+class ImageURL(BaseModel):
+    url: str
+
+
+class MultimodalInputItem(BaseModel):
+    type: Literal["text", "image_url"]
+    text: Optional[str] = None
+    image_url: Optional[ImageURL] = None
+
+
+class ChatMessage(BaseModel):
+    role: Role
+    content: Optional[Union[str, List[MultimodalInputItem]]] = None
+    tool_calls: Optional[List[FunctionCall]] = None
+
+
+class ChatCompletionMessage(BaseModel):
+    role: Optional[Role] = None
+    content: Optional[str] = None
+    tool_calls: Optional[List[FunctionCall]] = None
+
+
+class ChatCompletionRequest(BaseModel):
+    model: str
+    messages: List[ChatMessage]
+    tools: Optional[List[FunctionAvailable]] = None
+    do_sample: Optional[bool] = None
+    temperature: Optional[float] = None
+    top_p: Optional[float] = None
+    n: int = 1
+    max_tokens: Optional[int] = None
+    stop: Optional[Union[str, List[str]]] = None
+    stream: bool = False
+
+
+class ChatCompletionResponseChoice(BaseModel):
+    index: int
+    message: ChatCompletionMessage
+    finish_reason: Finish
+
+
+class ChatCompletionStreamResponseChoice(BaseModel):
+    index: int
+    delta: ChatCompletionMessage
+    finish_reason: Optional[Finish] = None
+
+
+class ChatCompletionResponseUsage(BaseModel):
+    prompt_tokens: int
+    completion_tokens: int
+    total_tokens: int
+
+
+class ChatCompletionResponse(BaseModel):
+    id: str
+    object: Literal["chat.completion"] = "chat.completion"
+    created: int = Field(default_factory=lambda: int(time.time()))
+    model: str
+    choices: List[ChatCompletionResponseChoice]
+    usage: ChatCompletionResponseUsage
+
+
+class ChatCompletionStreamResponse(BaseModel):
+    id: str
+    object: Literal["chat.completion.chunk"] = "chat.completion.chunk"
+    created: int = Field(default_factory=lambda: int(time.time()))
+    model: str
+    choices: List[ChatCompletionStreamResponseChoice]
+
+
+class ScoreEvaluationRequest(BaseModel):
+    model: str
+    messages: List[str]
+    max_length: Optional[int] = None
+
+
+class ScoreEvaluationResponse(BaseModel):
+    id: str
+    object: Literal["score.evaluation"] = "score.evaluation"
+    model: str
+    scores: List[float]

src/llamafactory/chat/__init__.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .base_engine import BaseEngine
+from .chat_model import ChatModel
+
+
+__all__ = ["BaseEngine", "ChatModel"]

src/llamafactory/chat/base_engine.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Any, AsyncGenerator, Dict, List, Literal, Optional, Sequence, Union
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedModel, PreTrainedTokenizer
+    from vllm import AsyncLLMEngine
+
+    from ..data import Template
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from ..hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
+
+
+@dataclass
+class Response:
+    response_text: str
+    response_length: int
+    prompt_length: int
+    finish_reason: Literal["stop", "length"]
+
+
+class BaseEngine(ABC):
+    r"""
+    Base class for inference engine of chat models.
+
+    Must implements async methods: chat(), stream_chat() and get_scores().
+    """
+
+    model: Union["PreTrainedModel", "AsyncLLMEngine"]
+    tokenizer: "PreTrainedTokenizer"
+    can_generate: bool
+    template: "Template"
+    generating_args: Dict[str, Any]
+
+    @abstractmethod
+    def __init__(
+        self,
+        model_args: "ModelArguments",
+        data_args: "DataArguments",
+        finetuning_args: "FinetuningArguments",
+        generating_args: "GeneratingArguments",
+    ) -> None:
+        r"""
+        Initializes an inference engine.
+        """
+        ...
+
+    @abstractmethod
+    async def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        r"""
+        Gets a list of responses of the chat model.
+        """
+        ...
+
+    @abstractmethod
+    async def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        r"""
+        Gets the response token-by-token of the chat model.
+        """
+        ...
+
+    @abstractmethod
+    async def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        r"""
+        Gets a list of scores of the reward model.
+        """
+        ...

src/llamafactory/chat/chat_model.py

+# Copyright 2024 THUDM and the LlamaFactory team.
+#
+# This code is inspired by the THUDM's ChatGLM implementation.
+# https://github.com/THUDM/ChatGLM-6B/blob/main/cli_demo.py
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+import os
+from threading import Thread
+from typing import TYPE_CHECKING, Any, AsyncGenerator, Dict, Generator, List, Optional, Sequence
+
+from ..extras.misc import torch_gc
+from ..hparams import get_infer_args
+from .hf_engine import HuggingfaceEngine
+from .vllm_engine import VllmEngine
+
+
+if TYPE_CHECKING:
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from .base_engine import BaseEngine, Response
+
+
+def _start_background_loop(loop: "asyncio.AbstractEventLoop") -> None:
+    asyncio.set_event_loop(loop)
+    loop.run_forever()
+
+
+class ChatModel:
+    r"""
+    General class for chat models. Backed by huggingface or vllm engines.
+
+    Supports both sync and async methods.
+    Sync methods: chat(), stream_chat() and get_scores().
+    Async methods: achat(), astream_chat() and aget_scores().
+    """
+
+    def __init__(self, args: Optional[Dict[str, Any]] = None) -> None:
+        model_args, data_args, finetuning_args, generating_args = get_infer_args(args)
+        self.engine_type = model_args.infer_backend
+        if model_args.infer_backend == "huggingface":
+            self.engine: "BaseEngine" = HuggingfaceEngine(model_args, data_args, finetuning_args, generating_args)
+        elif model_args.infer_backend == "vllm":
+            self.engine: "BaseEngine" = VllmEngine(model_args, data_args, finetuning_args, generating_args)
+        else:
+            raise NotImplementedError("Unknown backend: {}".format(model_args.infer_backend))
+
+        self._loop = asyncio.new_event_loop()
+        self._thread = Thread(target=_start_background_loop, args=(self._loop,), daemon=True)
+        self._thread.start()
+
+    def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        r"""
+        Gets a list of responses of the chat model.
+        """
+        task = asyncio.run_coroutine_threadsafe(
+            self.achat(messages, system, tools, image, video, **input_kwargs), self._loop
+        )
+        return task.result()
+
+    async def achat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        r"""
+        Asynchronously gets a list of responses of the chat model.
+        """
+        return await self.engine.chat(messages, system, tools, image, video, **input_kwargs)
+
+    def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> Generator[str, None, None]:
+        r"""
+        Gets the response token-by-token of the chat model.
+        """
+        generator = self.astream_chat(messages, system, tools, image, video, **input_kwargs)
+        while True:
+            try:
+                task = asyncio.run_coroutine_threadsafe(generator.__anext__(), self._loop)
+                yield task.result()
+            except StopAsyncIteration:
+                break
+
+    async def astream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        r"""
+        Asynchronously gets the response token-by-token of the chat model.
+        """
+        async for new_token in self.engine.stream_chat(messages, system, tools, image, video, **input_kwargs):
+            yield new_token
+
+    def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        r"""
+        Gets a list of scores of the reward model.
+        """
+        task = asyncio.run_coroutine_threadsafe(self.aget_scores(batch_input, **input_kwargs), self._loop)
+        return task.result()
+
+    async def aget_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        r"""
+        Asynchronously gets a list of scores of the reward model.
+        """
+        return await self.engine.get_scores(batch_input, **input_kwargs)
+
+
+def run_chat() -> None:
+    if os.name != "nt":
+        try:
+            import readline  # noqa: F401
+        except ImportError:
+            print("Install `readline` for a better experience.")
+
+    chat_model = ChatModel()
+    messages = []
+    print("Welcome to the CLI application, use `clear` to remove the history, use `exit` to exit the application.")
+
+    while True:
+        try:
+            query = input("\nUser: ")
+        except UnicodeDecodeError:
+            print("Detected decoding error at the inputs, please set the terminal encoding to utf-8.")
+            continue
+        except Exception:
+            raise
+
+        if query.strip() == "exit":
+            break
+
+        if query.strip() == "clear":
+            messages = []
+            torch_gc()
+            print("History has been removed.")
+            continue
+
+        messages.append({"role": "user", "content": query})
+        print("Assistant: ", end="", flush=True)
+
+        response = ""
+        for new_text in chat_model.stream_chat(messages):
+            print(new_text, end="", flush=True)
+            response += new_text
+        print()
+        messages.append({"role": "assistant", "content": response})

src/llamafactory/chat/hf_engine.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import asyncio
+import concurrent.futures
+import os
+from threading import Thread
+from typing import TYPE_CHECKING, Any, AsyncGenerator, Callable, Dict, List, Optional, Sequence, Tuple, Union
+
+import torch
+from transformers import GenerationConfig, TextIteratorStreamer
+from typing_extensions import override
+
+from ..data import get_template_and_fix_tokenizer
+from ..extras.constants import IMAGE_PLACEHOLDER, VIDEO_PLACEHOLDER
+from ..extras.logging import get_logger
+from ..extras.misc import get_logits_processor
+from ..model import load_model, load_tokenizer
+from .base_engine import BaseEngine, Response
+
+
+if TYPE_CHECKING:
+    from transformers import PreTrainedModel, PreTrainedTokenizer, ProcessorMixin
+    from trl import PreTrainedModelWrapper
+
+    from ..data import Template
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from ..hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
+
+
+logger = get_logger(__name__)
+
+
+class HuggingfaceEngine(BaseEngine):
+    def __init__(
+        self,
+        model_args: "ModelArguments",
+        data_args: "DataArguments",
+        finetuning_args: "FinetuningArguments",
+        generating_args: "GeneratingArguments",
+    ) -> None:
+        self.can_generate = finetuning_args.stage == "sft"
+        tokenizer_module = load_tokenizer(model_args)
+        self.tokenizer = tokenizer_module["tokenizer"]
+        self.processor = tokenizer_module["processor"]
+        self.tokenizer.padding_side = "left" if self.can_generate else "right"
+        self.template = get_template_and_fix_tokenizer(self.tokenizer, data_args)
+        self.model = load_model(
+            self.tokenizer, model_args, finetuning_args, is_trainable=False, add_valuehead=(not self.can_generate)
+        )  # must after fixing tokenizer to resize vocab
+        self.generating_args = generating_args.to_dict()
+        try:
+            asyncio.get_event_loop()
+        except RuntimeError:
+            logger.warning("There is no current event loop, creating a new one.")
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+
+        self.semaphore = asyncio.Semaphore(int(os.environ.get("MAX_CONCURRENT", "1")))
+
+    @staticmethod
+    def _process_args(
+        model: "PreTrainedModel",
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+        template: "Template",
+        generating_args: Dict[str, Any],
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> Tuple[Dict[str, Any], int]:
+        mm_input_dict = {"images": [], "videos": [], "imglens": [0], "vidlens": [0]}
+        if image is not None:
+            mm_input_dict.update({"images": [image], "imglens": [1]})
+            if IMAGE_PLACEHOLDER not in messages[0]["content"]:
+                messages[0]["content"] = IMAGE_PLACEHOLDER + messages[0]["content"]
+
+        if video is not None:
+            mm_input_dict.update({"videos": [video], "vidlens": [1]})
+            if VIDEO_PLACEHOLDER not in messages[0]["content"]:
+                messages[0]["content"] = VIDEO_PLACEHOLDER + messages[0]["content"]
+
+        messages = template.mm_plugin.process_messages(
+            messages, mm_input_dict["images"], mm_input_dict["videos"], processor
+        )
+        paired_messages = messages + [{"role": "assistant", "content": ""}]
+        system = system or generating_args["default_system"]
+        prompt_ids, _ = template.encode_oneturn(tokenizer, paired_messages, system, tools)
+        prompt_ids, _ = template.mm_plugin.process_token_ids(
+            prompt_ids, None, mm_input_dict["images"], mm_input_dict["videos"], tokenizer, processor
+        )
+        prompt_length = len(prompt_ids)
+        inputs = torch.tensor([prompt_ids], device=model.device)
+        attention_mask = torch.ones_like(inputs, dtype=torch.bool)
+
+        do_sample: Optional[bool] = input_kwargs.pop("do_sample", None)
+        temperature: Optional[float] = input_kwargs.pop("temperature", None)
+        top_p: Optional[float] = input_kwargs.pop("top_p", None)
+        top_k: Optional[float] = input_kwargs.pop("top_k", None)
+        num_return_sequences: int = input_kwargs.pop("num_return_sequences", 1)
+        repetition_penalty: Optional[float] = input_kwargs.pop("repetition_penalty", None)
+        length_penalty: Optional[float] = input_kwargs.pop("length_penalty", None)
+        max_length: Optional[int] = input_kwargs.pop("max_length", None)
+        max_new_tokens: Optional[int] = input_kwargs.pop("max_new_tokens", None)
+        stop: Optional[Union[str, List[str]]] = input_kwargs.pop("stop", None)
+
+        if stop is not None:
+            logger.warning("Stop parameter is not supported by the huggingface engine yet.")
+
+        generating_args = generating_args.copy()
+        generating_args.update(
+            dict(
+                do_sample=do_sample if do_sample is not None else generating_args["do_sample"],
+                temperature=temperature if temperature is not None else generating_args["temperature"],
+                top_p=top_p if top_p is not None else generating_args["top_p"],
+                top_k=top_k if top_k is not None else generating_args["top_k"],
+                num_return_sequences=num_return_sequences,
+                repetition_penalty=repetition_penalty
+                if repetition_penalty is not None
+                else generating_args["repetition_penalty"],
+                length_penalty=length_penalty if length_penalty is not None else generating_args["length_penalty"],
+                eos_token_id=[tokenizer.eos_token_id] + tokenizer.additional_special_tokens_ids,
+                pad_token_id=tokenizer.pad_token_id,
+            )
+        )
+
+        if isinstance(num_return_sequences, int) and num_return_sequences > 1:  # do_sample needs temperature > 0
+            generating_args["do_sample"] = True
+            generating_args["temperature"] = generating_args["temperature"] or 1.0
+
+        if not generating_args["temperature"]:
+            generating_args["do_sample"] = False
+
+        if not generating_args["do_sample"]:
+            generating_args.pop("temperature", None)
+            generating_args.pop("top_p", None)
+
+        if max_length:
+            generating_args.pop("max_new_tokens", None)
+            generating_args["max_length"] = max_length
+
+        if max_new_tokens:
+            generating_args.pop("max_length", None)
+            generating_args["max_new_tokens"] = max_new_tokens
+
+        gen_kwargs = dict(
+            inputs=inputs,
+            attention_mask=attention_mask,
+            generation_config=GenerationConfig(**generating_args),
+            logits_processor=get_logits_processor(),
+        )
+
+        mm_inputs = template.mm_plugin.get_mm_inputs(**mm_input_dict, seqlens=[prompt_length], processor=processor)
+        for key, value in mm_inputs.items():
+            value = value if isinstance(value, torch.Tensor) else torch.tensor(value)
+            gen_kwargs[key] = value.to(model.device)
+
+        return gen_kwargs, prompt_length
+
+    @staticmethod
+    @torch.inference_mode()
+    def _chat(
+        model: "PreTrainedModel",
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+        template: "Template",
+        generating_args: Dict[str, Any],
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> List["Response"]:
+        gen_kwargs, prompt_length = HuggingfaceEngine._process_args(
+            model, tokenizer, processor, template, generating_args, messages, system, tools, image, video, input_kwargs
+        )
+        generate_output = model.generate(**gen_kwargs)
+        response_ids = generate_output[:, prompt_length:]
+        response = tokenizer.batch_decode(response_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)
+        results = []
+        for i in range(len(response)):
+            eos_index = (response_ids[i] == tokenizer.eos_token_id).nonzero()
+            response_length = (eos_index[0].item() + 1) if len(eos_index) else len(response_ids[i])
+            results.append(
+                Response(
+                    response_text=response[i],
+                    response_length=response_length,
+                    prompt_length=prompt_length,
+                    finish_reason="stop" if len(eos_index) else "length",
+                )
+            )
+
+        return results
+
+    @staticmethod
+    @torch.inference_mode()
+    def _stream_chat(
+        model: "PreTrainedModel",
+        tokenizer: "PreTrainedTokenizer",
+        processor: Optional["ProcessorMixin"],
+        template: "Template",
+        generating_args: Dict[str, Any],
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> Callable[[], str]:
+        gen_kwargs, _ = HuggingfaceEngine._process_args(
+            model, tokenizer, processor, template, generating_args, messages, system, tools, image, video, input_kwargs
+        )
+        streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
+        gen_kwargs["streamer"] = streamer
+        thread = Thread(target=model.generate, kwargs=gen_kwargs, daemon=True)
+        thread.start()
+
+        def stream():
+            try:
+                return streamer.__next__()
+            except StopIteration:
+                raise StopAsyncIteration()
+
+        return stream
+
+    @staticmethod
+    @torch.inference_mode()
+    def _get_scores(
+        model: "PreTrainedModelWrapper",
+        tokenizer: "PreTrainedTokenizer",
+        batch_input: List[str],
+        input_kwargs: Optional[Dict[str, Any]] = {},
+    ) -> List[float]:
+        max_length: Optional[int] = input_kwargs.pop("max_length", None)
+        device = getattr(model.pretrained_model, "device", "cuda")
+        inputs: Dict[str, "torch.Tensor"] = tokenizer(
+            batch_input,
+            padding=True,
+            truncation=True,
+            max_length=max_length or getattr(model.config, "max_position_embeddings", 1024),
+            return_tensors="pt",
+            add_special_tokens=False,
+        ).to(device)
+        values: "torch.Tensor" = model(**inputs, return_dict=True, use_cache=False)[-1]
+        scores = values.gather(dim=-1, index=(inputs["attention_mask"].sum(dim=-1, keepdim=True) - 1))
+        return scores
+
+    @override
+    async def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        if not self.can_generate:
+            raise ValueError("The current model does not support `chat`.")
+
+        loop = asyncio.get_running_loop()
+        input_args = (
+            self.model,
+            self.tokenizer,
+            self.processor,
+            self.template,
+            self.generating_args,
+            messages,
+            system,
+            tools,
+            image,
+            video,
+            input_kwargs,
+        )
+        async with self.semaphore:
+            with concurrent.futures.ThreadPoolExecutor() as pool:
+                return await loop.run_in_executor(pool, self._chat, *input_args)
+
+    @override
+    async def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        if not self.can_generate:
+            raise ValueError("The current model does not support `stream_chat`.")
+
+        loop = asyncio.get_running_loop()
+        input_args = (
+            self.model,
+            self.tokenizer,
+            self.processor,
+            self.template,
+            self.generating_args,
+            messages,
+            system,
+            tools,
+            image,
+            video,
+            input_kwargs,
+        )
+        async with self.semaphore:
+            with concurrent.futures.ThreadPoolExecutor() as pool:
+                stream = self._stream_chat(*input_args)
+                while True:
+                    try:
+                        yield await loop.run_in_executor(pool, stream)
+                    except StopAsyncIteration:
+                        break
+
+    @override
+    async def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        if self.can_generate:
+            raise ValueError("Cannot get scores using an auto-regressive model.")
+
+        loop = asyncio.get_running_loop()
+        input_args = (self.model, self.tokenizer, batch_input, input_kwargs)
+        async with self.semaphore:
+            with concurrent.futures.ThreadPoolExecutor() as pool:
+                return await loop.run_in_executor(pool, self._get_scores, *input_args)

src/llamafactory/chat/vllm_engine.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import uuid
+from typing import TYPE_CHECKING, Any, AsyncGenerator, AsyncIterator, Dict, List, Optional, Sequence, Union
+
+from typing_extensions import override
+
+from ..data import get_template_and_fix_tokenizer
+from ..extras.constants import IMAGE_PLACEHOLDER
+from ..extras.logging import get_logger
+from ..extras.misc import get_device_count
+from ..extras.packages import is_pillow_available, is_vllm_available
+from ..model import load_config, load_tokenizer
+from ..model.model_utils.quantization import QuantizationMethod
+from ..model.model_utils.visual import LlavaMultiModalProjectorForYiVLForVLLM
+from .base_engine import BaseEngine, Response
+
+
+if is_pillow_available():
+    from PIL import Image
+    from PIL.Image import Image as ImageObject
+
+
+if is_vllm_available():
+    from vllm import AsyncEngineArgs, AsyncLLMEngine, RequestOutput, SamplingParams
+    from vllm.lora.request import LoRARequest
+
+
+if TYPE_CHECKING:
+    from ..data.mm_plugin import ImageInput, VideoInput
+    from ..hparams import DataArguments, FinetuningArguments, GeneratingArguments, ModelArguments
+
+
+logger = get_logger(__name__)
+
+
+class VllmEngine(BaseEngine):
+    def __init__(
+        self,
+        model_args: "ModelArguments",
+        data_args: "DataArguments",
+        finetuning_args: "FinetuningArguments",
+        generating_args: "GeneratingArguments",
+    ) -> None:
+        config = load_config(model_args)  # may download model from ms hub
+        if getattr(config, "quantization_config", None):  # gptq models should use float16
+            quantization_config: Dict[str, Any] = getattr(config, "quantization_config", None)
+            quant_method = quantization_config.get("quant_method", "")
+            if quant_method == QuantizationMethod.GPTQ and model_args.infer_dtype == "auto":
+                model_args.infer_dtype = "float16"
+
+        self.can_generate = finetuning_args.stage == "sft"
+        tokenizer_module = load_tokenizer(model_args)
+        self.tokenizer = tokenizer_module["tokenizer"]
+        self.processor = tokenizer_module["processor"]
+        self.tokenizer.padding_side = "left"
+        self.template = get_template_and_fix_tokenizer(self.tokenizer, data_args)
+        self.generating_args = generating_args.to_dict()
+
+        engine_args = {
+            "model": model_args.model_name_or_path,
+            "trust_remote_code": True,
+            "download_dir": model_args.cache_dir,
+            "dtype": model_args.infer_dtype,
+            "max_model_len": model_args.vllm_maxlen,
+            "tensor_parallel_size": get_device_count() or 1,
+            "gpu_memory_utilization": model_args.vllm_gpu_util,
+            "disable_log_stats": True,
+            "disable_log_requests": True,
+            "enforce_eager": model_args.vllm_enforce_eager,
+            "enable_lora": model_args.adapter_name_or_path is not None,
+            "max_lora_rank": model_args.vllm_max_lora_rank,
+        }
+
+        if getattr(config, "is_yi_vl_derived_model", None):
+            import vllm.model_executor.models.llava
+
+            logger.info("Detected Yi-VL model, applying projector patch.")
+            vllm.model_executor.models.llava.LlavaMultiModalProjector = LlavaMultiModalProjectorForYiVLForVLLM
+
+        self.model = AsyncLLMEngine.from_engine_args(AsyncEngineArgs(**engine_args))
+        if model_args.adapter_name_or_path is not None:
+            self.lora_request = LoRARequest("default", 1, model_args.adapter_name_or_path[0])
+        else:
+            self.lora_request = None
+
+    async def _generate(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncIterator["RequestOutput"]:
+        request_id = "chatcmpl-{}".format(uuid.uuid4().hex)
+        if image is not None:
+            if IMAGE_PLACEHOLDER not in messages[0]["content"]:
+                messages[0]["content"] = IMAGE_PLACEHOLDER + messages[0]["content"]
+
+        paired_messages = messages + [{"role": "assistant", "content": ""}]
+        system = system or self.generating_args["default_system"]
+        prompt_ids, _ = self.template.encode_oneturn(self.tokenizer, paired_messages, system, tools)
+        prompt_length = len(prompt_ids)
+
+        use_beam_search: bool = self.generating_args["num_beams"] > 1
+        temperature: Optional[float] = input_kwargs.pop("temperature", None)
+        top_p: Optional[float] = input_kwargs.pop("top_p", None)
+        top_k: Optional[float] = input_kwargs.pop("top_k", None)
+        num_return_sequences: int = input_kwargs.pop("num_return_sequences", 1)
+        repetition_penalty: Optional[float] = input_kwargs.pop("repetition_penalty", None)
+        length_penalty: Optional[float] = input_kwargs.pop("length_penalty", None)
+        max_length: Optional[int] = input_kwargs.pop("max_length", None)
+        max_new_tokens: Optional[int] = input_kwargs.pop("max_new_tokens", None)
+        stop: Optional[Union[str, List[str]]] = input_kwargs.pop("stop", None)
+
+        if "max_new_tokens" in self.generating_args:
+            max_tokens = self.generating_args["max_new_tokens"]
+        elif "max_length" in self.generating_args:
+            if self.generating_args["max_length"] > prompt_length:
+                max_tokens = self.generating_args["max_length"] - prompt_length
+            else:
+                max_tokens = 1
+
+        if max_length:
+            max_tokens = max_length - prompt_length if max_length > prompt_length else 1
+
+        if max_new_tokens:
+            max_tokens = max_new_tokens
+
+        sampling_params = SamplingParams(
+            n=num_return_sequences,
+            repetition_penalty=(
+                repetition_penalty if repetition_penalty is not None else self.generating_args["repetition_penalty"]
+            )
+            or 1.0,  # repetition_penalty must > 0
+            temperature=temperature if temperature is not None else self.generating_args["temperature"],
+            top_p=(top_p if top_p is not None else self.generating_args["top_p"]) or 1.0,  # top_p must > 0
+            top_k=top_k if top_k is not None else self.generating_args["top_k"],
+            use_beam_search=use_beam_search,
+            length_penalty=length_penalty if length_penalty is not None else self.generating_args["length_penalty"],
+            stop=stop,
+            stop_token_ids=[self.tokenizer.eos_token_id] + self.tokenizer.additional_special_tokens_ids,
+            max_tokens=max_tokens,
+            skip_special_tokens=True,
+        )
+
+        if image is not None:  # add image features
+            if not isinstance(image, (str, ImageObject)):
+                raise ValueError("Expected image input is a path or PIL.Image, but got {}.".format(type(image)))
+
+            if isinstance(image, str):
+                image = Image.open(image).convert("RGB")
+
+            multi_modal_data = {"image": image}
+        else:
+            multi_modal_data = None
+
+        result_generator = self.model.generate(
+            inputs={"prompt_token_ids": prompt_ids, "multi_modal_data": multi_modal_data},
+            sampling_params=sampling_params,
+            request_id=request_id,
+            lora_request=self.lora_request,
+        )
+        return result_generator
+
+    @override
+    async def chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> List["Response"]:
+        final_output = None
+        generator = await self._generate(messages, system, tools, image, video, **input_kwargs)
+        async for request_output in generator:
+            final_output = request_output
+
+        results = []
+        for output in final_output.outputs:
+            results.append(
+                Response(
+                    response_text=output.text,
+                    response_length=len(output.token_ids),
+                    prompt_length=len(final_output.prompt_token_ids),
+                    finish_reason=output.finish_reason,
+                )
+            )
+
+        return results
+
+    @override
+    async def stream_chat(
+        self,
+        messages: Sequence[Dict[str, str]],
+        system: Optional[str] = None,
+        tools: Optional[str] = None,
+        image: Optional["ImageInput"] = None,
+        video: Optional["VideoInput"] = None,
+        **input_kwargs,
+    ) -> AsyncGenerator[str, None]:
+        generated_text = ""
+        generator = await self._generate(messages, system, tools, image, video, **input_kwargs)
+        async for result in generator:
+            delta_text = result.outputs[0].text[len(generated_text) :]
+            generated_text = result.outputs[0].text
+            yield delta_text
+
+    @override
+    async def get_scores(
+        self,
+        batch_input: List[str],
+        **input_kwargs,
+    ) -> List[float]:
+        raise NotImplementedError("vLLM engine does not support get_scores.")

src/llamafactory/cli.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import random
+import subprocess
+import sys
+from enum import Enum, unique
+
+from . import launcher
+from .api.app import run_api
+from .chat.chat_model import run_chat
+from .eval.evaluator import run_eval
+from .extras.env import VERSION, print_env
+from .extras.logging import get_logger
+from .extras.misc import get_device_count
+from .train.tuner import export_model, run_exp
+from .webui.interface import run_web_demo, run_web_ui
+
+
+USAGE = (
+    "-" * 70
+    + "\n"
+    + "| Usage:                                                             |\n"
+    + "|   llamafactory-cli api -h: launch an OpenAI-style API server       |\n"
+    + "|   llamafactory-cli chat -h: launch a chat interface in CLI         |\n"
+    + "|   llamafactory-cli eval -h: evaluate models                        |\n"
+    + "|   llamafactory-cli export -h: merge LoRA adapters and export model |\n"
+    + "|   llamafactory-cli train -h: train models                          |\n"
+    + "|   llamafactory-cli webchat -h: launch a chat interface in Web UI   |\n"
+    + "|   llamafactory-cli webui: launch LlamaBoard                        |\n"
+    + "|   llamafactory-cli version: show version info                      |\n"
+    + "-" * 70
+)
+
+WELCOME = (
+    "-" * 58
+    + "\n"
+    + "| Welcome to LLaMA Factory, version {}".format(VERSION)
+    + " " * (21 - len(VERSION))
+    + "|\n|"
+    + " " * 56
+    + "|\n"
+    + "| Project page: https://github.com/hiyouga/LLaMA-Factory |\n"
+    + "-" * 58
+)
+
+logger = get_logger(__name__)
+
+
+@unique
+class Command(str, Enum):
+    API = "api"
+    CHAT = "chat"
+    ENV = "env"
+    EVAL = "eval"
+    EXPORT = "export"
+    TRAIN = "train"
+    WEBDEMO = "webchat"
+    WEBUI = "webui"
+    VER = "version"
+    HELP = "help"
+
+
+def main():
+    command = sys.argv.pop(1) if len(sys.argv) != 1 else Command.HELP
+    if command == Command.API:
+        run_api()
+    elif command == Command.CHAT:
+        run_chat()
+    elif command == Command.ENV:
+        print_env()
+    elif command == Command.EVAL:
+        run_eval()
+    elif command == Command.EXPORT:
+        export_model()
+    elif command == Command.TRAIN:
+        force_torchrun = os.environ.get("FORCE_TORCHRUN", "0").lower() in ["true", "1"]
+        if force_torchrun or get_device_count() > 1:
+            master_addr = os.environ.get("MASTER_ADDR", "127.0.0.1")
+            master_port = os.environ.get("MASTER_PORT", str(random.randint(20001, 29999)))
+            logger.info("Initializing distributed tasks at: {}:{}".format(master_addr, master_port))
+            process = subprocess.run(
+                (
+                    "torchrun --nnodes {nnodes} --node_rank {node_rank} --nproc_per_node {nproc_per_node} "
+                    "--master_addr {master_addr} --master_port {master_port} {file_name} {args}"
+                ).format(
+                    nnodes=os.environ.get("NNODES", "1"),
+                    node_rank=os.environ.get("RANK", "0"),
+                    nproc_per_node=os.environ.get("NPROC_PER_NODE", str(get_device_count())),
+                    master_addr=master_addr,
+                    master_port=master_port,
+                    file_name=launcher.__file__,
+                    args=" ".join(sys.argv[1:]),
+                ),
+                shell=True,
+            )
+            sys.exit(process.returncode)
+        else:
+            run_exp()
+    elif command == Command.WEBDEMO:
+        run_web_demo()
+    elif command == Command.WEBUI:
+        run_web_ui()
+    elif command == Command.VER:
+        print(WELCOME)
+    elif command == Command.HELP:
+        print(USAGE)
+    else:
+        raise NotImplementedError("Unknown command: {}.".format(command))

src/llamafactory/data/__init__.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from .collator import (
+    KTODataCollatorWithPadding,
+    MultiModalDataCollatorForSeq2Seq,
+    PairwiseDataCollatorWithPadding,
+    SFTDataCollatorWith4DAttentionMask,
+)
+from .data_utils import Role, split_dataset
+from .loader import get_dataset
+from .template import TEMPLATES, Template, get_template_and_fix_tokenizer
+
+
+__all__ = [
+    "KTODataCollatorWithPadding",
+    "MultiModalDataCollatorForSeq2Seq",
+    "PairwiseDataCollatorWithPadding",
+    "SFTDataCollatorWith4DAttentionMask",
+    "Role",
+    "split_dataset",
+    "get_dataset",
+    "TEMPLATES",
+    "Template",
+    "get_template_and_fix_tokenizer",
+]

src/llamafactory/data/aligner.py

+# Copyright 2024 the LlamaFactory team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+from functools import partial
+from typing import TYPE_CHECKING, Any, Dict, List, Optional, Sequence, Union
+
+from ..extras.logging import get_logger
+from .data_utils import Role
+
+
+if TYPE_CHECKING:
+    from datasets import Dataset, IterableDataset
+    from transformers import Seq2SeqTrainingArguments
+
+    from ..hparams import DataArguments
+    from .mm_plugin import ImageInput, VideoInput
+    from .parser import DatasetAttr
+
+
+logger = get_logger(__name__)
+
+
+def _convert_images(
+    images: Sequence["ImageInput"],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Optional[List["ImageInput"]]:
+    r"""
+    Optionally concatenates image path to dataset dir when loading from local disk.
+    """
+    if len(images) == 0:
+        return None
+
+    images = images[:]
+    if dataset_attr.load_from in ["script", "file"]:
+        for i in range(len(images)):
+            if isinstance(images[i], str) and os.path.isfile(os.path.join(data_args.dataset_dir, images[i])):
+                images[i] = os.path.join(data_args.dataset_dir, images[i])
+
+    return images
+
+
+def _convert_videos(
+    videos: Sequence["VideoInput"],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Optional[List["VideoInput"]]:
+    r"""
+    Optionally concatenates video path to dataset dir when loading from local disk.
+    """
+    if len(videos) == 0:
+        return None
+
+    videos = videos[:]
+    if dataset_attr.load_from in ["script", "file"]:
+        for i in range(len(videos)):
+            if isinstance(videos[i], str) and os.path.isfile(os.path.join(data_args.dataset_dir, videos[i])):
+                videos[i] = os.path.join(data_args.dataset_dir, videos[i])
+
+    return videos
+
+
+def convert_alpaca(
+    example: Dict[str, Any],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Dict[str, Any]:
+    r"""
+    Converts alpaca format dataset to the standard format.
+    """
+    prompt = []
+    if dataset_attr.history and isinstance(example[dataset_attr.history], list):
+        for old_prompt, old_response in example[dataset_attr.history]:
+            prompt.append({"role": Role.USER.value, "content": old_prompt})
+            prompt.append({"role": Role.ASSISTANT.value, "content": old_response})
+
+    query = []
+    if dataset_attr.prompt and example[dataset_attr.prompt]:
+        query.append(example[dataset_attr.prompt])
+
+    if dataset_attr.query and example[dataset_attr.query]:
+        query.append(example[dataset_attr.query])
+
+    prompt.append({"role": Role.USER.value, "content": "\n".join(query)})  # "prompt\nquery"
+
+    if dataset_attr.kto_tag and isinstance(example[dataset_attr.kto_tag], bool):  # kto example
+        response = [{"role": Role.ASSISTANT.value, "content": example[dataset_attr.response]}]
+        if example[dataset_attr.kto_tag]:
+            response = response + [{"role": Role.ASSISTANT.value, "content": ""}]
+        else:
+            response = [{"role": Role.ASSISTANT.value, "content": ""}] + response
+    elif (
+        dataset_attr.ranking
+        and isinstance(example[dataset_attr.chosen], str)
+        and isinstance(example[dataset_attr.rejected], str)
+    ):  # pairwise example
+        response = [
+            {"role": Role.ASSISTANT.value, "content": example[dataset_attr.chosen]},
+            {"role": Role.ASSISTANT.value, "content": example[dataset_attr.rejected]},
+        ]
+    elif dataset_attr.response and isinstance(example[dataset_attr.response], str):  # normal example
+        response = [{"role": Role.ASSISTANT.value, "content": example[dataset_attr.response]}]
+    else:  # unsupervised
+        response = []
+
+    convert_images = partial(_convert_images, dataset_attr=dataset_attr, data_args=data_args)
+    convert_videos = partial(_convert_videos, dataset_attr=dataset_attr, data_args=data_args)
+    output = {
+        "_prompt": prompt,
+        "_response": response,
+        "_system": example[dataset_attr.system] if dataset_attr.system else "",
+        "_tools": example[dataset_attr.tools] if dataset_attr.tools else "",
+        "_images": convert_images(example[dataset_attr.images]) if dataset_attr.images else None,
+        "_videos": convert_videos(example[dataset_attr.videos]) if dataset_attr.videos else None,
+    }
+    return output
+
+
+def convert_sharegpt(
+    example: Dict[str, Any],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+) -> Dict[str, Any]:
+    r"""
+    Converts sharegpt format dataset to the standard format.
+    """
+    tag_mapping = {
+        dataset_attr.user_tag: Role.USER.value,
+        dataset_attr.assistant_tag: Role.ASSISTANT.value,
+        dataset_attr.observation_tag: Role.OBSERVATION.value,
+        dataset_attr.function_tag: Role.FUNCTION.value,
+        dataset_attr.system_tag: Role.SYSTEM.value,
+    }
+    odd_tags = (dataset_attr.user_tag, dataset_attr.observation_tag)
+    even_tags = (dataset_attr.assistant_tag, dataset_attr.function_tag)
+    accept_tags = (odd_tags, even_tags)
+    messages = example[dataset_attr.messages]
+    if (
+        dataset_attr.system_tag
+        and len(messages) != 0
+        and messages[0][dataset_attr.role_tag] == dataset_attr.system_tag
+    ):
+        system = messages[0][dataset_attr.content_tag]
+        messages = messages[1:]
+    else:
+        system = example[dataset_attr.system] if dataset_attr.system else ""
+
+    aligned_messages = []
+    broken_data = False
+    for turn_idx, message in enumerate(messages):
+        if message[dataset_attr.role_tag] not in accept_tags[turn_idx % 2]:
+            logger.warning("Invalid role tag in {}.".format(messages))
+            broken_data = True
+
+        aligned_messages.append(
+            {"role": tag_mapping[message[dataset_attr.role_tag]], "content": message[dataset_attr.content_tag]}
+        )
+
+    if (not dataset_attr.ranking and len(aligned_messages) % 2 != 0) or (
+        dataset_attr.ranking and len(aligned_messages) % 2 == 0
+    ):
+        logger.warning("Invalid message count in {}.".format(messages))
+        broken_data = True
+
+    if dataset_attr.kto_tag and isinstance(example[dataset_attr.kto_tag], bool):  # kto example
+        prompt = aligned_messages[:-1]
+        response = aligned_messages[-1:]
+        if example[dataset_attr.kto_tag]:
+            response = response + [{"role": Role.ASSISTANT.value, "content": ""}]
+        else:
+            response = [{"role": Role.ASSISTANT.value, "content": ""}] + response
+    elif (
+        dataset_attr.ranking
+        and isinstance(example[dataset_attr.chosen], dict)
+        and isinstance(example[dataset_attr.rejected], dict)
+    ):  # pairwise example
+        chosen = example[dataset_attr.chosen]
+        rejected = example[dataset_attr.rejected]
+        if (
+            chosen[dataset_attr.role_tag] not in accept_tags[-1]
+            or rejected[dataset_attr.role_tag] not in accept_tags[-1]
+        ):
+            logger.warning("Invalid role tag in {}.".format([chosen, rejected]))
+            broken_data = True
+
+        prompt = aligned_messages
+        response = [
+            {"role": tag_mapping[chosen[dataset_attr.role_tag]], "content": chosen[dataset_attr.content_tag]},
+            {"role": tag_mapping[rejected[dataset_attr.role_tag]], "content": rejected[dataset_attr.content_tag]},
+        ]
+    else:  # normal example
+        prompt = aligned_messages[:-1]
+        response = aligned_messages[-1:]
+
+    if broken_data:
+        logger.warning("Skipping this abnormal example.")
+        prompt, response = [], []
+
+    convert_images = partial(_convert_images, dataset_attr=dataset_attr, data_args=data_args)
+    convert_videos = partial(_convert_videos, dataset_attr=dataset_attr, data_args=data_args)
+    output = {
+        "_prompt": prompt,
+        "_response": response,
+        "_system": system,
+        "_tools": example[dataset_attr.tools] if dataset_attr.tools else "",
+        "_images": convert_images(example[dataset_attr.images]) if dataset_attr.images else None,
+        "_videos": convert_videos(example[dataset_attr.videos]) if dataset_attr.videos else None,
+    }
+    return output
+
+
+def align_dataset(
+    dataset: Union["Dataset", "IterableDataset"],
+    dataset_attr: "DatasetAttr",
+    data_args: "DataArguments",
+    training_args: "Seq2SeqTrainingArguments",
+) -> Union["Dataset", "IterableDataset"]:
+    r"""
+    Aligned dataset:
+        _prompt: [{"role": "user", "content": "..."}] * (2T - 1)
+        _response: [{"role": "assistant", "content": "..."}] * N (N > 1 for ranking dataset)
+        _system: "..."
+        _tools: "...",
+        _images: [],
+        _videos: [],
+    """
+    if dataset_attr.formatting == "alpaca":
+        convert_func = partial(convert_alpaca, dataset_attr=dataset_attr, data_args=data_args)
+    else:
+        convert_func = partial(convert_sharegpt, dataset_attr=dataset_attr, data_args=data_args)
+
+    column_names = list(next(iter(dataset)).keys())
+    kwargs = {}
+    if not data_args.streaming:
+        kwargs = dict(
+            num_proc=data_args.preprocessing_num_workers,
+            load_from_cache_file=(not data_args.overwrite_cache) or (training_args.local_process_index != 0),
+            desc="Converting format of dataset",
+        )
+
+    return dataset.map(
+        convert_func,
+        batched=False,
+        remove_columns=column_names,
+        **kwargs,
+    )