feat(translator): 添加transformers作为备选模型加载方式
支持使用transformers库作为llama-cpp-python的备选方案加载模型 新增模型加载失败时的自动回退机制 更新requirements.txt添加transformers和torch依赖
This commit is contained in:
7
models/Modelfile
Normal file
7
models/Modelfile
Normal file
@@ -0,0 +1,7 @@
|
||||
FROM ./HY-MT1.5-1.8B_bf16_Q4_K_M.gguf
|
||||
|
||||
PARAMETER temperature 0.7
|
||||
PARAMETER top_p 0.95
|
||||
PARAMETER num_ctx 2048
|
||||
|
||||
SYSTEM 你是一个专业的翻译助手，根据以下要求将中文翻译成英文。
|
||||
@@ -3,4 +3,6 @@ llama-cpp-python
|
||||
python-docx
|
||||
loguru
|
||||
psutil
|
||||
GPUtil
|
||||
GPUtil
|
||||
transformers
|
||||
torch
|
||||
129
translator.py
129
translator.py
@@ -6,10 +6,18 @@ try:
|
||||
from llama_cpp import Llama
|
||||
llama_cpp_available = True
|
||||
except ImportError:
|
||||
logger.warning("llama-cpp-python库未找到，将禁用翻译功能")
|
||||
logger.warning("llama-cpp-python库未找到，将尝试使用transformers库")
|
||||
Llama = None
|
||||
llama_cpp_available = False
|
||||
|
||||
# 尝试导入transformers库
|
||||
try:
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
transformers_available = True
|
||||
except ImportError:
|
||||
logger.warning("transformers库未找到")
|
||||
transformers_available = False
|
||||
|
||||
class Translator:
|
||||
def __init__(self, model_path=None):
|
||||
self.model = None
|
||||
@@ -17,13 +25,11 @@ class Translator:
|
||||
self.is_ready = False
|
||||
self.model_name = ""
|
||||
self.llama_cpp_available = llama_cpp_available
|
||||
self.transformers_available = transformers_available
|
||||
self.use_transformers = False
|
||||
|
||||
def load_model(self, model_path=None):
|
||||
"""加载模型"""
|
||||
if not self.llama_cpp_available:
|
||||
logger.error("llama-cpp-python库未找到，无法加载模型")
|
||||
return False
|
||||
|
||||
if model_path:
|
||||
self.model_path = model_path
|
||||
|
||||
@@ -31,22 +37,73 @@ class Translator:
|
||||
logger.error("未提供模型路径")
|
||||
return False
|
||||
|
||||
if not os.path.exists(self.model_path):
|
||||
logger.error(f"模型文件不存在: {self.model_path}")
|
||||
return False
|
||||
|
||||
try:
|
||||
logger.info(f"开始加载模型: {self.model_path}")
|
||||
self.model = Llama(
|
||||
model_path=self.model_path,
|
||||
n_ctx=2048,
|
||||
n_threads=4,
|
||||
n_gpu_layers=100 # 尽可能使用GPU加速
|
||||
)
|
||||
self.is_ready = True
|
||||
self.model_name = os.path.basename(self.model_path)
|
||||
logger.info(f"模型加载成功: {self.model_name}")
|
||||
return True
|
||||
|
||||
if self.llama_cpp_available:
|
||||
if os.path.exists(self.model_path):
|
||||
try:
|
||||
self.model = Llama(
|
||||
model_path=self.model_path,
|
||||
n_ctx=2048,
|
||||
n_threads=4,
|
||||
n_gpu_layers=100
|
||||
)
|
||||
self.use_transformers = False
|
||||
self.is_ready = True
|
||||
self.model_name = os.path.basename(self.model_path)
|
||||
logger.info(f"模型加载成功: {self.model_name}")
|
||||
return True
|
||||
except Exception as e:
|
||||
logger.warning(f"使用llama-cpp-python加载模型失败: {e}")
|
||||
else:
|
||||
logger.warning(f"模型文件不存在: {self.model_path}")
|
||||
|
||||
if self.transformers_available:
|
||||
try:
|
||||
from transformers import AutoConfig
|
||||
if os.path.exists(self.model_path):
|
||||
config = AutoConfig.from_pretrained(
|
||||
self.model_path,
|
||||
local_files_only=True,
|
||||
trust_remote_code=True
|
||||
)
|
||||
self.model = AutoModelForCausalLM.from_pretrained(
|
||||
self.model_path,
|
||||
local_files_only=True,
|
||||
trust_remote_code=True,
|
||||
torch_dtype="auto"
|
||||
)
|
||||
self.tokenizer = AutoTokenizer.from_pretrained(
|
||||
self.model_path,
|
||||
local_files_only=True,
|
||||
trust_remote_code=True
|
||||
)
|
||||
else:
|
||||
config = AutoConfig.from_pretrained(
|
||||
self.model_path,
|
||||
trust_remote_code=True
|
||||
)
|
||||
self.model = AutoModelForCausalLM.from_pretrained(
|
||||
self.model_path,
|
||||
trust_remote_code=True,
|
||||
torch_dtype="auto"
|
||||
)
|
||||
self.tokenizer = AutoTokenizer.from_pretrained(
|
||||
self.model_path,
|
||||
trust_remote_code=True
|
||||
)
|
||||
self.use_transformers = True
|
||||
self.is_ready = True
|
||||
self.model_name = self.model_path
|
||||
logger.info(f"使用transformers加载模型成功: {self.model_name}")
|
||||
return True
|
||||
except Exception as e:
|
||||
logger.error(f"使用transformers加载模型失败: {e}")
|
||||
return False
|
||||
|
||||
logger.error("没有可用的模型加载方式")
|
||||
return False
|
||||
except Exception as e:
|
||||
logger.error(f"模型加载失败: {e}")
|
||||
self.is_ready = False
|
||||
@@ -54,10 +111,6 @@ class Translator:
|
||||
|
||||
def translate(self, text, context="", terms=None):
|
||||
"""执行翻译"""
|
||||
if not self.llama_cpp_available:
|
||||
logger.error("llama-cpp-python库未找到，无法执行翻译")
|
||||
return ""
|
||||
|
||||
if not self.is_ready or not self.model:
|
||||
logger.error("模型未就绪，无法执行翻译")
|
||||
return ""
|
||||
@@ -68,16 +121,28 @@ class Translator:
|
||||
|
||||
logger.info(f"开始翻译，输入长度: {len(text)} 字符")
|
||||
|
||||
# 调用模型进行翻译
|
||||
output = self.model(
|
||||
prompt,
|
||||
max_tokens=2048,
|
||||
temperature=0.7,
|
||||
top_p=0.95,
|
||||
stop=["\n原文:", "\n译文:", "\n###"]
|
||||
)
|
||||
if self.use_transformers:
|
||||
import torch
|
||||
inputs = self.tokenizer(prompt, return_tensors="pt")
|
||||
with torch.no_grad():
|
||||
outputs = self.model.generate(
|
||||
**inputs,
|
||||
max_new_tokens=2048,
|
||||
temperature=0.7,
|
||||
top_p=0.95,
|
||||
do_sample=True
|
||||
)
|
||||
translated_text = self.tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
|
||||
else:
|
||||
output = self.model(
|
||||
prompt,
|
||||
max_tokens=2048,
|
||||
temperature=0.7,
|
||||
top_p=0.95,
|
||||
stop=["\n原文:", "\n译文:", "\n###"]
|
||||
)
|
||||
translated_text = output["choices"][0]["text"].strip()
|
||||
|
||||
translated_text = output["choices"][0]["text"].strip()
|
||||
logger.info(f"翻译完成，输出长度: {len(translated_text)} 字符")
|
||||
|
||||
return translated_text
|
||||
|
||||
Reference in New Issue
Block a user