torch 1.13
cuda 11.7

推理代码能正常运行

训练开4卡4090，加载模型时卡住，cpu占用100%，显卡占用100%

torchrun --nnodes=1 --nproc_per_node=4 --rdzv_id=1 --rdzv_backend=c10d --rdzv_endpoint=localhost:12345 finetune_cpm_bee.py --use-delta --model-config config/cpm-bee-10b.json --dataset datasets/eprstmt/binary/dev --eval_dataset datasets/eprstmt/binary/eval_dev --epoch 100 --batch-size 4 --train-iters 100 --save-name cpm_bee_finetune --max-length 2048 --save results/ --lr 0.0001 --inspect-iters 100 --warmup-iters 1 --eval-interval 1000 --early-stop-patience 5 --lr-decay-style noam --weight-decay 0.01 --clip-grad 1.0 --loss-scale 32768 --start-step 0 --load path/pytorch_model_10b.bin

====================== Initialization ======================
rank : 0
local_rank : 0
world_size : 4
local_size : 4
master : star-SYS-420GP-TNR:37257
device : 0
cpus : [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 1
3, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 2
4, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 3
5, 36, 37]

Traceback (most recent call last):
File "finetune_cpm_bee.py", line 4, in
import bmtrain as bmt
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 2, in
from .init import init_distributed
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 8, in
from . import nccl
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/init.py", line 4, in
from . import _C as C
ImportError: /home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/_C.cpython-38-x86_64-linux-gnu.so: undefined symbol: ncclBroadcast
Traceback (most recent call last):
File "finetune_cpm_bee.py", line 4, in
import bmtrain as bmt
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 2, in
from .init import init_distributed
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 8, in
from . import nccl
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/init.py", line 4, in
from . import _C as C
ImportError: /home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/_C.cpython-38-x86_64-linux-gnu.so: undefined symbol: ncclBroadcast
Traceback (most recent call last):
File "finetune_cpm_bee.py", line 4, in
import bmtrain as bmt
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 2, in
from .init import init_distributed
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 8, in
from . import nccl
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/init.py", line 4, in
from . import _C as C
ImportError: /home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/_C.cpython-38-x86_64-linux-gnu.so: undefined symbol: ncclBroadcast
Traceback (most recent call last):
File "finetune_cpm_bee.py", line 4, in
import bmtrain as bmt
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 2, in
from .init import init_distributed
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/init.py", line 8, in
from . import nccl
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/init.py", line 4, in
from . import _C as C
ImportError: /home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/bmtrain/nccl/_C.cpython-38-x86_64-linux-gnu.so: undefined symbol: ncclBroadcast
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 81486) of binary: /home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/bin/python
Traceback (most recent call last):
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/bin/torchrun", line 11, in
sys.exit(main())
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/torch/distributed/elastic/multiprocessing/errors/init.py", line 346, in wrapper
return f(*args, **kwargs)
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/torch/distributed/run.py", line 762, in main
run(args)
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/torch/distributed/run.py", line 753, in run
elastic_launch(
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 132, in call
return launch_agent(self._config, self._entrypoint, list(args))
File "/home/mdisk2/tanjunwen/anaconda3/envs/cpmbee2/lib/python3.8/site-packages/torch/distributed/launcher/api.py", line 246, in launch_agent
raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:

Dear Editor,
All of you did a good job in the LLM's exploration.
I try to use CPM-Bee-2B , But I find that requesting in Chinese will return English results,
while requesting in English will return Chinese results.
It is so bad fo me :(
there is my data
data_list =   
[{"input": "我生病了,写一份请假单", "prompt": "内容符合规范", "": ""},
    {"input": "I’m sick, write a leave of absence form.", "prompt": "make enough word", "": ""}]
But i get a bad odd answer.

I know the model too small.
Did I enter the language in the wrong format?
Hope we can have a talk about this, Thank u.
Yours,

请问如果想要在CPM-Bee模型上同时微调几个不同的子任务，数据格式需要设置成什么样的？谢谢，盼复

discord/wechat/qq/clack?

load model /models/cpm-bee-10b/pytorch_model.bin please wait ......
sucess loaded model /models/cpm-bee-10b/pytorch_model.bin
Enter json line to evaluate: { "prompt": "Java冒泡排序算法"}
Error: ''
press Enter to continue...
{ "prompt": "Java冒泡排序算法","": ""}
Enter json line to evaluate: { "prompt": "Java冒泡排序算法","": ""}
{'prompt': 'Java冒泡排序算法', '': 'Java冒泡排序算法'}
press Enter to continue...

Enter json line to evaluate: {"input": "我生病了,写一份请假单", "prompt": "内容符合规范", "": ""}
Error: ''
press Enter to continue...

Enter json line to evaluate: {"input": "我生病了,写一份请假单", "prompt": "内容符合规范", "": ""}
{'input': '我生病了,写一份请假单', 'prompt': '内容符合规范', '': ',老板不批假怎么办?'}
Enter json line to evaluate: {"input": "购物车服务有添加商品，删除商品，购物车结算等功能", "prompt": "写代码", "": ""}
{'input': '购物车服务有添加商品，删除商品，购物车结算等功能', 'prompt': '写代码', '': '。\n购物车服务的核心是购物车结算，购物车结算有两种方式：\n1、先将商品添加到购物车里面，然后再进行结算；\n2、直接在购物车里面就可以进行结算。'}

Enter json line to evaluate: {"document": "今天天气很好，我和妈妈一起去公园，", "prompt": "往后写约100字", "": ""}
{'document': '今天天气很好，我和妈妈一起去公园，', 'prompt': '往后写约100字', '': '公园里有很多人，我和妈妈一起去玩滑梯了。\n今天天气很好，我和妈妈一起去公园，公园里有很多人，我和妈妈一起去玩滑梯了。\n今天天气很好，我和妈妈一起去公园，公园里有很多人，我和妈妈一起去玩滑梯了。\n今天天气很好，我和妈妈一起去公园，公园里有很多人。'}

比如代码生成， 逻辑推理，文本生成，都明显比之前清华6b的差。

下面是改进的代码，可以从控制台输入，方便测试，希望纳入代码库中帮助别人，谢谢。

from cpm_live.generation.bee import CPMBeeBeamSearch
from cpm_live.models import CPMBeeTorch, CPMBeeConfig
from cpm_live.tokenizers import CPMBeeTokenizer
from opendelta import LoraModel
import torch
import json

if name == "main":

data_list = [

{"document": "今天天气是真的<mask_0>", "": {"<mask_0>": ""}},
{"input": "今天天气很好，我和妈妈一起去公园，", "prompt": "往后写约100字", "": ""},
{"input": "北京是**的首都", "prompt": "中翻英", "": ""},
{"input": "NGC 6231是一个位于天蝎座的疏散星团，天球座标为赤经16时54分，赤纬-41度48分，", "question": "NGC 6231的经纬度是多少？", "": ""},
{"input":"之前多次聚餐都选择这里，现在由于炭火改成了电烤羊，口感真的不如从前，","question":"评分是多少？(1-5)","":""},
{"input": "父母都希望自己的孩子诚实、勇敢、有礼貌。父母首先得从自己做起，", "options": {"<option_0>": "少提要求", "<option_1>": "降低标准", "<option_2>": "自己先做好", "<option_3>": "让孩子拿主意"}, "question": "教育孩子时，父母应该：", "": ""}
]
print('sample input data ')
for data in data_list :
print(json.dumps(data,ensure_ascii=False))
config = CPMBeeConfig.from_json_file("config/cpm-bee-10b.json")
ckpt_path = "/models/cpm-bee-10b/pytorch_model.bin"
tokenizer = CPMBeeTokenizer()
model = CPMBeeTorch(config=config)

# insert LoRA if your model has been finetuned in delta-tuning.
# delta_model = LoraModel(backbone_model=model, modified_modules=["project_q", "project_v"], backend="hf")
# lora_ckpt_path = "path/to/lora.pt"
#model.load_state_dict(torch.load(lora_ckpt_path), strict=False)
print('load model '+ckpt_path +" please wait ......")
model.load_state_dict(torch.load(ckpt_path), strict=False)
model.cuda()
print('sucess loaded model '+ckpt_path)
while True:
   # Read input from console
   line = input("Enter json line to evaluate: ")
   json_array =[ json.loads(line)]
   #print(json_array)
  # Evaluate input as a Python expression
   try:
      # use beam search
      beam_search = CPMBeeBeamSearch(
      model=model,
      tokenizer=tokenizer,
      )
      inference_results = beam_search.generate(json_array , max_length=100, repetition_penalty=1.1)
      for res in inference_results:
         print(res)
   except Exception as e:
     result = f"Error: {e}"
     print(result)

   # Wait for continued input
   input("press Enter to continue...\n")

您好，请问未来是否会开源评测代码？

克隆了$ git clone -b main --single-branch https://github.com/OpenBMB/CPM-Bee.git
装了库：pip install -r requirements.txt
转换了数据
但是微调没跑通，报错如下：

╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ /data/rain-llm/code/CPM-Bee/src/finetune_cpm_bee.py:193 in <module>                              │
│                                                                                                  │
│   190 │   model: CPMBee,                                                                         │
│   191 │   optimizer: bmt.optim.AdamOffloadOptimizer,                                             │
│   192 │   lr_scheduler: bmt.lr_scheduler.WarmupLRScheduler,                                      │
│ ❱ 193 │   optim_manager: bmt.optim.OptimManager,                                                 │
│   194 ):                                                                                         │
│   195 │                                                                                          │
│   196 │   average_time = bmt.utils.AverageRecorder()                                             │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
AttributeError: module 'bmtrain.optim' has no attribute 'OptimManager'
╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ /data/rain-llm/code/CPM-Bee/src/finetune_cpm_bee.py:193 in <module>                              │
│                                                                                                  │
│   190 │   model: CPMBee,                                                                         │
│   191 │   optimizer: bmt.optim.AdamOffloadOptimizer,                                             │
│   192 │   lr_scheduler: bmt.lr_scheduler.WarmupLRScheduler,                                      │
│ ❱ 193 │   optim_manager: bmt.optim.OptimManager,                                                 │
│   194 ):                                                                                         │
│   195 │                                                                                          │
│   196 │   average_time = bmt.utils.AverageRecorder()                                             │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
AttributeError: module 'bmtrain.optim' has no attribute 'OptimManager'
╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ /data/rain-llm/code/CPM-Bee/src/finetune_cpm_bee.py:193 in <module>                              │
│                                                                                                  │
│   190 │   model: CPMBee,                                                                         │
│   191 │   optimizer: bmt.optim.AdamOffloadOptimizer,                                             │
│   192 │   lr_scheduler: bmt.lr_scheduler.WarmupLRScheduler,                                      │
│ ❱ 193 │   optim_manager: bmt.optim.OptimManager,                                                 │
│   194 ):                                                                                         │
│   195 │                                                                                          │
│   196 │   average_time = bmt.utils.AverageRecorder()                                             │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
AttributeError: module 'bmtrain.optim' has no attribute 'OptimManager'
╭─────────────────────────────── Traceback (most recent call last) ────────────────────────────────╮
│ /data/rain-llm/code/CPM-Bee/src/finetune_cpm_bee.py:193 in <module>                              │
│                                                                                                  │
│   190 │   model: CPMBee,                                                                         │
│   191 │   optimizer: bmt.optim.AdamOffloadOptimizer,                                             │
│   192 │   lr_scheduler: bmt.lr_scheduler.WarmupLRScheduler,                                      │
│ ❱ 193 │   optim_manager: bmt.optim.OptimManager,                                                 │
│   194 ):                                                                                         │
│   195 │                                                                                          │
│   196 │   average_time = bmt.utils.AverageRecorder()                                             │
╰──────────────────────────────────────────────────────────────────────────────────────────────────╯
AttributeError: module 'bmtrain.optim' has no attribute 'OptimManager'
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 94429) of binary: /opt/conda/bin/python3
Traceback (most recent call last):
  File "/opt/conda/bin/torchrun", line 33, in <module>
    sys.exit(load_entry_point('torch==2.0.0', 'console_scripts', 'torchrun')())
  File "/opt/conda/lib/python3.10/site-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 346, in wrapper
    return f(*args, **kwargs)
  File "/opt/conda/lib/python3.10/site-packages/torch/distributed/run.py", line 794, in main
    run(args)
  File "/opt/conda/lib/python3.10/site-packages/torch/distributed/run.py", line 785, in run
    elastic_launch(
  File "/opt/conda/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 134, in __call__
    return launch_agent(self._config, self._entrypoint, list(args))
  File "/opt/conda/lib/python3.10/site-packages/torch/distributed/launcher/api.py", line 250, in launch_agent
    raise ChildFailedError(
torch.distributed.elastic.multiprocessing.errors.ChildFailedError: 
============================================================
finetune_cpm_bee.py FAILED
------------------------------------------------------------
Failures:
[1]:
  time      : 2023-05-28_22:57:29
  host      : ubuntu
  rank      : 1 (local_rank: 1)
  exitcode  : 1 (pid: 94430)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
[2]:
  time      : 2023-05-28_22:57:29
  host      : ubuntu
  rank      : 2 (local_rank: 2)
  exitcode  : 1 (pid: 94431)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
[3]:
  time      : 2023-05-28_22:57:29
  host      : ubuntu
  rank      : 3 (local_rank: 3)
  exitcode  : 1 (pid: 94432)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html
------------------------------------------------------------
Root Cause (first observed failure):
[0]:
  time      : 2023-05-28_22:57:29
  host      : ubuntu
  rank      : 0 (local_rank: 0)
  exitcode  : 1 (pid: 94429)
  error_file: <N/A>
  traceback : To enable traceback see: https://pytorch.org/docs/stable/elastic/errors.html

直接运行推理没有报错

python text_generation.py

{'document': '**最好的大学是<mask_0>', '': {'<mask_0>': '清华北大，世界最好的大学是麻省理工。'}}

Dear authors,

Thanks for your contribution to the community!!! :-) The OpenBMB series is awesome!

I wonder if a technical report will be released to disclose some training and evaluation/benchmark details.

Thanks and have a good day!

Best,
Zhihong

请问 BMTrain 现在是否能够适配CUDA 12

Traceback (most recent call last):
File "cpmbee_translator.py", line 2, in
from cpm_live.generation.bee import CPMBeeBeamSearch
File "/home/css/CPM-Bee/src/cpm_live/generation/init.py", line 1, in
from .ant import CPMAntBeamSearch, CPMAntRandomSampling, CPMAntGeneration
File "/home/css/CPM-Bee/src/cpm_live/generation/ant.py", line 4, in
from ..utils import pad
File "/home/css/CPM-Bee/src/cpm_live/utils/init.py", line 1, in
from .config import Config
File "/home/css/CPM-Bee/src/cpm_live/utils/config.py", line 20, in
from .log import logger
File "/home/css/CPM-Bee/src/cpm_live/utils/log.py", line 27, in
logger = _get_logger()
File "/home/css/CPM-Bee/src/cpm_live/utils/log.py", line 15, in _get_logger
node_name = os.getenv("NODE_NAME", str(bmt.rank()))
File "/opt/conda/lib/python3.8/site-packages/bmtrain-0.0.15-py3.8-linux-x86_64.egg/bmtrain/global_var.py", line 24, in rank
return config['rank']
KeyError: 'rank'

运行demo直接报该错误，请问有什么方法解决吗，尝试在 text_generation.py中加入以下代码也不行
import os
os.environ['MASTER_ADDR'] = 'localhost'
os.environ['MASTER_PORT'] = '5678'

!pip install -r requirements.txt
Looking in indexes: https://pypi.org/simple, https://us-python.pkg.dev/colab-wheels/public/simple/
Collecting torch<2.0.0,>=1.10
Using cached torch-1.13.1-cp310-cp310-manylinux1_x86_64.whl (887.5 MB)
Collecting bmtrain>=0.2.1
Using cached bmtrain-0.2.2.tar.gz (58 kB)
error: subprocess-exited-with-error

× python setup.py egg_info did not run successfully.
│ exit code: 1
╰─> See above for output.

note: This error originates from a subprocess, and is likely not a problem with pip.
Preparing metadata (setup.py) ... error
error: metadata-generation-failed

× Encountered error while generating package metadata.
╰─> See above for output.

note: This is an issue with the package mentioned above, not pip.
hint: See above for details.

有关数据格式的更多信息，您可以查看CPM-Bee数据格式

CPM-Bee数据格式在哪？

如题，使用预处理数据脚本预处理数据后，在进行微调时，dataset报错，不太确定是预处理脚本的问题，还是dataset迭代的问题

376 if self._max_repeat_times is not None:
377 if self._repeat_times >= self._max_repeat_times:
--> 378 raise EOFError("End of dataset")
379 print('_prepare_new_epoch, 2')
380 nw_unused_block: List[int] = []

finetune的时候如果load参数为 预训练模型pytorch_model.bin文件，损失就为nan；如果不加载预训练模型pytorch_model.bin文件那就是从头开始训了？

运行下面的代码会报错，经过测试是因为含有"<"
"""
from cpm_live.models import CPMBeeTorch, CPMBeeConfig
from cpm_live.tokenizers import CPMBeeTokenizer
config = CPMBeeConfig.from_json_file("config/cpm-bee-10b.json")
tokenizer = CPMBeeTokenizer()
print(tokenizer._special_tokens)
text = "if 成绩 < 60"
tokens = tokenizer.tokenize(text)
"""

File "text_generation.py", line 28, in
tokens = tokenizer.tokenize(text)
File "/root/CPM-Bee/src/cpm_live/tokenizers/bee.py", line 143, in tokenize
raise ValueError("Unexpected end of text {}".format(text))
ValueError: Unexpected end of text if 成绩 < 60

windowns部署的时候，编译bmtrain不过，缺少nccl

请问是否有技术交流群呢？方便大家一起沟通一下。

在api中，问答使用的是question，生成使用的是prompt，区别在哪里呢?

when running text_generation, bmtrain error appeared:
ImportError: /home/CPM-Bee/BMTrain/bmtrain/optim/_cuda.cpython-39-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops9new_zeros4callERKNS_6TensorEN3c108ArrayRefINS5_6SymIntEEENS5_8optionalINS5_10ScalarTypeEEENS9_INS5_6LayoutEEENS9_INS5_6DeviceEEENS9_IbEE
can you please provide conda env or Dockerfile?

例子给出 ，ckpt_path = "cpm-bee-3b-ckpt.pt" 这个模型文件从哪里下载

在生成文本时，怎么能做到流式输出呢？

我尝试用了{"input": "问题：xxx答案：", "": "yyy"} 做预训练，但是loss基本上为nan。但是对于微调是正确的，请教正确的预训练格式是什么

是否支持langchain的调用，能否提供相应的文档

I have download the model cpm-bee-10b from huggingface ,but I do not understand the path of /your/model/checkpointin.
Can you someone please guide me to solve this problem?

`
RuntimeError:
The detected CUDA version (12.1) mismatches the version that was used to compile
PyTorch (11.8). Please make sure to use the same CUDA versions.

我的电脑上的cuda版本nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Mon_Apr__3_17:16:06_PDT_2023
Cuda compilation tools, release 12.1, V12.1.105
Build cuda_12.1.r12.1/compiler.32688072_0
`
这台电脑上还有其它项目，难道，要我为了这一个项目降版本么，
能否修懒得说那里让这个项目支持12可能性

没有N卡，是否考虑支持一下Apple Silicon M1/M2 芯片呢？或者摩尔线程的显卡？

我看到README里面给出了6种特定任务的数据格式，想问一下如果要做通用的对话（假定就是单轮的），数据格式应该是怎样的呢？如果用文本生成的那种格式，prompt应该用什么比较好？我尝试了一些好像效果不太好

8卡机器，使用4，5，6，7这4块显卡；
默认安装的opendelta == 0.3.0
torch==2.0.0
transformers==4.28.1

全参数微调时报错如下：
TypeError: CheckpointBlock._named_members() got an unexpected keyword argument 'remove_duplicate'
ERROR:torch.distributed.elastic.multiprocessing.api:failed (exitcode: 1) local_rank: 0 (pid: 8973) of binary: /opt/conda/bin/python

I'm trying the demo code, here is the information: with CUDA 12.1

the command !python -c "import torch;print(torch.cuda.nccl.version())", can return (2, 14, 3)

below is the original import error stack:

---

ImportError Traceback (most recent call last)
Cell In[10], line 1
----> 1 from cpm_live.generation.bee import CPMBeeBeamSearch
2 from cpm_live.models import CPMBeeTorch, CPMBeeConfig
3 from cpm_live.tokenizers import CPMBeeTokenizer

File /workspace/cpm_live/generation/init.py:1
----> 1 from .ant import CPMAntBeamSearch, CPMAntRandomSampling, CPMAntGeneration

File /workspace/cpm_live/generation/ant.py:4
2 import torch.nn.functional as F
3 from .generation_utils import BeamHypotheses, apply_repetition_penalty, top_k_top_p_filtering
----> 4 from ..utils import pad
7 class CPMAntGeneration:
8 def init(self, model, tokenizer, prompt_length=32):

File /workspace/cpm_live/utils/init.py:1
----> 1 from .config import Config
2 from .data_utils import pad
3 from .object import allgather_objects

File /workspace/cpm_live/utils/config.py:20
18 import copy
19 from typing import Any, Dict, Union
---> 20 from .log import logger
23 def load_dataset_config(dataset_path: str):
24 cfg = json.load(open(dataset_path, "r", encoding="utf-8"))

File /workspace/cpm_live/utils/log.py:7
5 import json
6 import logging
----> 7 import bmtrain as bmt
10 # Set up the common logger
11 def _get_logger():

File /usr/local/lib/python3.10/dist-packages/bmtrain/init.py:2
1 from .global_var import config, world_size, rank
----> 2 from .init import init_distributed
4 from .parameter import DistributedParameter, ParameterInitializer
5 from .layer import DistributedModule

File /usr/local/lib/python3.10/dist-packages/bmtrain/init.py:8
6 from .utils import print_dict
7 from .global_var import config
----> 8 from . import nccl
9 from .synchronize import synchronize
10 def init_distributed(
11 init_method : str = "env://",
12 seed : int = 0,
(...)
15 num_micro_batches: int = None,
16 ):

File /usr/local/lib/python3.10/dist-packages/bmtrain/nccl/init.py:4
2 from typing_extensions import Literal
3 import torch
----> 4 from . import _C as C
5 from .enums import *
7 class NCCLCommunicator:

ImportError: /usr/local/lib/python3.10/dist-packages/bmtrain/nccl/_C.cpython-310-x86_64-linux-gnu.so: undefined symbol: ncclBroadcast

请问怎样加载增量微调后的模型，README中只有

delta_model = LoraModel(backbone_model=model, modified_modules=["project_q", "project_v"], backend="hf")

这句代码，没有给出加载模型权重的代码，求指导

请问如何获取整句的向量，做文本相似度或者文本分类

多轮对话格式如何定义？
使用一下这种格式是否可行？

下载使用：https://github.com/WangRongsheng/Use-LLMs-in-Colab/tree/main/CPM-Bee
在线使用：https://colab.research.google.com/drive/1w7J80aSgahCXFTEsqbZIjjo7ccAuGXhx?usp=sharing

如题

报错如下：
nccl.obj : error LNK2001: 无法解析的外部符号 ncclCommInitRank
。。。
build\lib.win-amd64-cpython-39\bmtrain\nccl_C.cp39-win_amd64.pyd : fatal error LNK1120: 15 个无法解析的外部命令。
感谢。

示例代码是将模型加载在单张显卡上推理的，如何将模型权重分配在多张卡上，有像hugging face那样简单的device_map='auto'的方法吗？

微调10B模型有推荐的GPU要求么，比如说至少多少内存l? 20G内存和单张3090是推理的需求吧，如果要跑finetune 10B模型，最少要几张3090呀, 支持QLora, ptuning这种节约计算的形式么?

do you have method besides 'pip install bmtrain'?

or give a example for install sucessfully, with what kinds of packages should be installed in advanced?