我要评论目录
前言
想必有小伙伴也想跟我一样体验下部署大语言模型, 但碍于经济实力, 不过民间上出现了大量的量化模型, 我们平民也能体验体验啦~, 该模型可以在笔记本电脑上部署, 确保你电脑至少有16g运行内存
开原地址：github - ymcui/chinese-llama-alpaca: 中文llama&alpaca大语言模型+本地cpu部署 (chinese llama & alpaca llms)
linux和mac的教程在开源的仓库中有提供，当然如果你是m1的也可以参考以下文章：
https://gist.github.com/cedrickchee/e8d4cb0c4b1df6cc47ce8b18457ebde0
准备工作
最好是有代理, 不然你下载东西可能失败, 我为了下个模型花了一天时间, 痛哭~
我们需要先在电脑上安装以下环境：
- git
- python3.9（使用anaconda3创建该环境）
- cmake（如果你电脑没有c和c++的编译环境还需要安装mingw）
git
下载地址：git - downloading package
在cmd窗口输入以下如果有版本号显示说明已经安装成功
git -v
python3.9
我这里使用anaconda3来使用python, anaconda3是什么？
anaconda3下载地址：anaconda | anaconda distribution
安装步骤参考：
在cmd窗口输入以下命令, 显示版本号则说明安装成功
conda -v
接下来我们在cmd窗口输入以下命令创建一个python3.9的环境
conda create --name py39 python=3.9 -y
查看有哪些环境的命令：
conda info -e
激活/切换环境的命令：
conda activate py39要使用哪个环境的话换成对应名字即可
进入环境后你就可以在这输入python相关的命令了, 如：
要退出环境的话输入：
conda deactivate当我退出环境后再查看python版本的话会提示我不是内部或外部命令，也不是可运行的程序
或批处理文件。如：
cmake
这是一个编译工具, 我们需要使用它去编译llama.cpp, 量化模型需要用到, 不量化模型个人电脑跑不起来, 觉得量化这个概念不理解的可以理解为压缩, 这种概念是不对的, 只是为了帮助你更好的理解.
在安装之前我们需要安装mingw, 避免编译时找不到编译环境, 按下win+r快捷键输入powershell
输入命令安装scoop, 这是一个包管理器, 我们使用它来下载安装mingw：
这个地方如果没有开代理的话可能会出错
iex "& {$(irm get.scoop.sh)} -runasadmin"安装好后分别运行下面两个命令（添加库）：
scoop bucket add extrasscoop bucket add main输入命令安装mingw
scoop install mingw到这就已经安装好mingw了, 如果报错了请评论, 我看到了会回复
接下来安装cmake
安装参考：
下载模型
我们需要下载两个模型, 一个是原版的llama模型, 一个是扩充了中文的模型, 后续会进行一个合并模型的操作
- 原版模型下载地址（要代理）：https://ipfs.io/ipfs/qmb9y5gcktg7zzbbwmu2bxwmkzyckcujtekppgdz7gefkm/
- 备用：nyanko7/llama-7b at main
- 扩充了中文的模型下载：
建议在d盘上新建一个文件夹, 在里面进行下载操作, 如下：
在弹出的框中分别输入以下命令：
git lfs installgit clone https://huggingface.co/ziqingyang/chinese-alpaca-lora-7b这里可能会因为网络问题一直失败......一直重试就行, 有别的问题请评论, 看到会回复
合并模型
终于写到这里了, 累~
在你下载了模型的目录内打开cmd窗口, 如下：
打开窗口后需要先激活python环境, 使用的就是前面装anaconda3
# 不记得有哪些环境的先运行以下命令
conda info -e
# 然后激活你需要的环境 我的环境名是py39
conda activate py39切换好后分别执行以下命令安装依赖库
pip install git+https://github.com/huggingface/transformers
pip install sentencepiece==0.1.97
pip install peft==0.2.0执行命令安装成功后会有successfully的字眼
接下来需要将原版模型转hf格式, 需要借助最新版🤗transformers提供的脚本convert_llama_weights_to_hf.py
在目录内新建一个convert_llama_weights_to_hf.py文件, 用记事本打开后把以下代码粘贴进去
注意：我这里是为了方便直接拷贝出来了，脚本可能会更新，建议直接去以下地址拷贝最新的：
transformers/convert_llama_weights_to_hf.py at main · huggingface/transformers · github
# copyright 2022 eleutherai and the huggingface inc. team. all rights reserved.
#
# 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 argparse
import gc
import json
import math
import os
import shutil
import warnings
import torch
from transformers import llamaconfig, llamaforcausallm, llamatokenizer
try:
from transformers import llamatokenizerfast
except importerror as e:
warnings.warn(e)
warnings.warn(
"the converted tokenizer will be the `slow` tokenizer. to use the fast, update your `tokenizers` library and re-run the tokenizer conversion"
)
llamatokenizerfast = none
"""
sample usage:
```
python src/transformers/models/llama/convert_llama_weights_to_hf.py \
--input_dir /path/to/downloaded/llama/weights --model_size 7b --output_dir /output/path
```
thereafter, models can be loaded via:
```py
from transformers import llamaforcausallm, llamatokenizer
model = llamaforcausallm.from_pretrained("/output/path")
tokenizer = llamatokenizer.from_pretrained("/output/path")
```
important note: you need to be able to host the whole model in ram to execute this script (even if the biggest versions
come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in ram).
"""
intermediate_size_map = {
"7b": 11008,
"13b": 13824,
"30b": 17920,
"65b": 22016,
}
num_shards = {
"7b": 1,
"13b": 2,
"30b": 4,
"65b": 8,
}
def compute_intermediate_size(n):
return int(math.ceil(n * 8 / 3) + 255) // 256 * 256
def read_json(path):
with open(path, "r") as f:
return json.load(f)
def write_json(text, path):
with open(path, "w") as f:
json.dump(text, f)
def write_model(model_path, input_base_path, model_size):
os.makedirs(model_path, exist_ok=true)
tmp_model_path = os.path.join(model_path, "tmp")
os.makedirs(tmp_model_path, exist_ok=true)
params = read_json(os.path.join(input_base_path, "params.json"))
num_shards = num_shards[model_size]
n_layers = params["n_layers"]
n_heads = params["n_heads"]
n_heads_per_shard = n_heads // num_shards
dim = params["dim"]
dims_per_head = dim // n_heads
base = 10000.0
inv_freq = 1.0 / (base ** (torch.arange(0, dims_per_head, 2).float() / dims_per_head))
# permute for sliced rotary
def permute(w):
return w.view(n_heads, dim // n_heads // 2, 2, dim).transpose(1, 2).reshape(dim, dim)
print(f"fetching all parameters from the checkpoint at {input_base_path}.")
# load weights
if model_size == "7b":
# not shared
# (the sharded implementation would also work, but this is simpler.)
loaded = torch.load(os.path.join(input_base_path, "consolidated.00.pth"), map_location="cpu")
else:
# sharded
loaded = [
torch.load(os.path.join(input_base_path, f"consolidated.{i:02d}.pth"), map_location="cpu")
for i in range(num_shards)
]
param_count = 0
index_dict = {"weight_map": {}}
for layer_i in range(n_layers):
filename = f"pytorch_model-{layer_i + 1}-of-{n_layers + 1}.bin"
if model_size == "7b":
# unsharded
state_dict = {
f"model.layers.{layer_i}.self_attn.q_proj.weight": permute(
loaded[f"layers.{layer_i}.attention.wq.weight"]
),
f"model.layers.{layer_i}.self_attn.k_proj.weight": permute(
loaded[f"layers.{layer_i}.attention.wk.weight"]
),
f"model.layers.{layer_i}.self_attn.v_proj.weight": loaded[f"layers.{layer_i}.attention.wv.weight"],
f"model.layers.{layer_i}.self_attn.o_proj.weight": loaded[f"layers.{layer_i}.attention.wo.weight"],
f"model.layers.{layer_i}.mlp.gate_proj.weight": loaded[f"layers.{layer_i}.feed_forward.w1.weight"],
f"model.layers.{layer_i}.mlp.down_proj.weight": loaded[f"layers.{layer_i}.feed_forward.w2.weight"],
f"model.layers.{layer_i}.mlp.up_proj.weight": loaded[f"layers.{layer_i}.feed_forward.w3.weight"],
f"model.layers.{layer_i}.input_layernorm.weight": loaded[f"layers.{layer_i}.attention_norm.weight"],
f"model.layers.{layer_i}.post_attention_layernorm.weight": loaded[f"layers.{layer_i}.ffn_norm.weight"],
}
else:
# sharded
# note that in the 13b checkpoint, not cloning the two following weights will result in the checkpoint
# becoming 37gb instead of 26gb for some reason.
state_dict = {
f"model.layers.{layer_i}.input_layernorm.weight": loaded[0][
f"layers.{layer_i}.attention_norm.weight"
].clone(),
f"model.layers.{layer_i}.post_attention_layernorm.weight": loaded[0][
f"layers.{layer_i}.ffn_norm.weight"
].clone(),
}
state_dict[f"model.layers.{layer_i}.self_attn.q_proj.weight"] = permute(
torch.cat(
[
loaded[i][f"layers.{layer_i}.attention.wq.weight"].view(n_heads_per_shard, dims_per_head, dim)
for i in range(num_shards)
],
dim=0,
).reshape(dim, dim)
)
state_dict[f"model.layers.{layer_i}.self_attn.k_proj.weight"] = permute(
torch.cat(
[
loaded[i][f"layers.{layer_i}.attention.wk.weight"].view(n_heads_per_shard, dims_per_head, dim)
for i in range(num_shards)
],
dim=0,
).reshape(dim, dim)
)
state_dict[f"model.layers.{layer_i}.self_attn.v_proj.weight"] = torch.cat(
[
loaded[i][f"layers.{layer_i}.attention.wv.weight"].view(n_heads_per_shard, dims_per_head, dim)
for i in range(num_shards)
],
dim=0,
).reshape(dim, dim)
state_dict[f"model.layers.{layer_i}.self_attn.o_proj.weight"] = torch.cat(
[loaded[i][f"layers.{layer_i}.attention.wo.weight"] for i in range(num_shards)], dim=1
)
state_dict[f"model.layers.{layer_i}.mlp.gate_proj.weight"] = torch.cat(
[loaded[i][f"layers.{layer_i}.feed_forward.w1.weight"] for i in range(num_shards)], dim=0
)
state_dict[f"model.layers.{layer_i}.mlp.down_proj.weight"] = torch.cat(
[loaded[i][f"layers.{layer_i}.feed_forward.w2.weight"] for i in range(num_shards)], dim=1
)
state_dict[f"model.layers.{layer_i}.mlp.up_proj.weight"] = torch.cat(
[loaded[i][f"layers.{layer_i}.feed_forward.w3.weight"] for i in range(num_shards)], dim=0
)
state_dict[f"model.layers.{layer_i}.self_attn.rotary_emb.inv_freq"] = inv_freq
for k, v in state_dict.items():
index_dict["weight_map"][k] = filename
param_count += v.numel()
torch.save(state_dict, os.path.join(tmp_model_path, filename))
filename = f"pytorch_model-{n_layers + 1}-of-{n_layers + 1}.bin"
if model_size == "7b":
# unsharded
state_dict = {
"model.embed_tokens.weight": loaded["tok_embeddings.weight"],
"model.norm.weight": loaded["norm.weight"],
"lm_head.weight": loaded["output.weight"],
}
else:
state_dict = {
"model.norm.weight": loaded[0]["norm.weight"],
"model.embed_tokens.weight": torch.cat(
[loaded[i]["tok_embeddings.weight"] for i in range(num_shards)], dim=1
),
"lm_head.weight": torch.cat([loaded[i]["output.weight"] for i in range(num_shards)], dim=0),
}
for k, v in state_dict.items():
index_dict["weight_map"][k] = filename
param_count += v.numel()
torch.save(state_dict, os.path.join(tmp_model_path, filename))
# write configs
index_dict["metadata"] = {"total_size": param_count * 2}
write_json(index_dict, os.path.join(tmp_model_path, "pytorch_model.bin.index.json"))
config = llamaconfig(
hidden_size=dim,
intermediate_size=compute_intermediate_size(dim),
num_attention_heads=params["n_heads"],
num_hidden_layers=params["n_layers"],
rms_norm_eps=params["norm_eps"],
)
config.save_pretrained(tmp_model_path)
# make space so we can load the model properly now.
del state_dict
del loaded
gc.collect()
print("loading the checkpoint in a llama model.")
model = llamaforcausallm.from_pretrained(tmp_model_path, torch_dtype=torch.float16, low_cpu_mem_usage=true)
# avoid saving this as part of the config.
del model.config._name_or_path
print("saving in the transformers format.")
model.save_pretrained(model_path)
shutil.rmtree(tmp_model_path)
def write_tokenizer(tokenizer_path, input_tokenizer_path):
# initialize the tokenizer based on the `spm` model
tokenizer_class = llamatokenizer if llamatokenizerfast is none else llamatokenizerfast
print("saving a {tokenizer_class} to {tokenizer_path}")
tokenizer = tokenizer_class(input_tokenizer_path)
tokenizer.save_pretrained(tokenizer_path)
def main():
parser = argparse.argumentparser()
parser.add_argument(
"--input_dir",
help="location of llama weights, which contains tokenizer.model and model folders",
)
parser.add_argument(
"--model_size",
choices=["7b", "13b", "30b", "65b", "tokenizer_only"],
)
parser.add_argument(
"--output_dir",
help="location to write hf model and tokenizer",
)
args = parser.parse_args()
if args.model_size != "tokenizer_only":
write_model(
model_path=args.output_dir,
input_base_path=os.path.join(args.input_dir, args.model_size),
model_size=args.model_size,
)
spm_path = os.path.join(args.input_dir, "tokenizer.model")
write_tokenizer(args.output_dir, spm_path)
if __name__ == "__main__":
main()在cmd窗口执行命令（如果你使用了anaconda，执行命令前请先激活环境）：
python convert_llama_weights_to_hf.py --input_dir path_to_original_llama_root_dir --model_size 7b --output_dir path_to_original_llama_hf_dir经过漫长的等待....
接下来合并输出pytorch版本权重（.pth文件），使用merge_llama_with_chinese_lora.py脚本
在目录新建一个merge_llama_with_chinese_lora.py文件, 用记事本打开将以下代码粘贴进去
注意：我这里是为了方便直接拷贝出来了，脚本可能会更新，建议直接去以下地址拷贝最新的：
chinese-llama-alpaca/merge_llama_with_chinese_lora.py at main · ymcui/chinese-llama-alpaca · github
"""
borrowed and modified from https://github.com/tloen/alpaca-lora
"""
import argparse
import os
import json
import gc
import torch
import transformers
import peft
from peft import peftmodel
parser = argparse.argumentparser()
parser.add_argument('--base_model',default=none,required=true,type=str,help="please specify a base_model")
parser.add_argument('--lora_model',default=none,required=true,type=str,help="please specify a lora_model")
# deprecated; the script infers the model size from the checkpoint
parser.add_argument('--model_size',default='7b',type=str,help="size of the llama model",choices=['7b','13b'])
parser.add_argument('--offload_dir',default=none,type=str,help="(optional) please specify a temp folder for offloading (useful for low-ram machines). default none (disable offload).")
parser.add_argument('--output_dir',default='./',type=str)
args = parser.parse_args()
assert (
"llamatokenizer" in transformers._import_structure["models.llama"]
), "llama is now in huggingface's main branch.\nplease reinstall it: pip uninstall transformers && pip install git+https://github.com/huggingface/transformers.git"
from transformers import llamatokenizer, llamaforcausallm
base_model = args.base_model
lora_model = args.lora_model
output_dir = args.output_dir
assert (
base_model
), "please specify a base_model in the script, e.g. 'decapoda-research/llama-7b-hf'"
tokenizer = llamatokenizer.from_pretrained(lora_model)
if args.offload_dir is not none:
# load with offloading, which is useful for low-ram machines.
# note that if you have enough ram, please use original method instead, as it is faster.
base_model = llamaforcausallm.from_pretrained(
base_model,
load_in_8bit=false,
torch_dtype=torch.float16,
offload_folder=args.offload_dir,
offload_state_dict=true,
low_cpu_mem_usage=true,
device_map={"": "cpu"},
)
else:
# original method without offloading
base_model = llamaforcausallm.from_pretrained(
base_model,
load_in_8bit=false,
torch_dtype=torch.float16,
device_map={"": "cpu"},
)
base_model.resize_token_embeddings(len(tokenizer))
assert base_model.get_input_embeddings().weight.size(0) == len(tokenizer)
tokenizer.save_pretrained(output_dir)
print(f"extended vocabulary size: {len(tokenizer)}")
first_weight = base_model.model.layers[0].self_attn.q_proj.weight
first_weight_old = first_weight.clone()
## infer the model size from the checkpoint
emb_to_model_size = {
4096 : '7b',
5120 : '13b',
6656 : '30b',
8192 : '65b',
}
embedding_size = base_model.get_input_embeddings().weight.size(1)
model_size = emb_to_model_size[embedding_size]
print(f"loading lora for {model_size} model")
lora_model = peftmodel.from_pretrained(
base_model,
lora_model,
device_map={"": "cpu"},
torch_dtype=torch.float16,
)
assert torch.allclose(first_weight_old, first_weight)
# merge weights
print(f"peft version: {peft.__version__}")
print(f"merging model")
if peft.__version__ > '0.2.0':
# merge weights - new merging method from peft
lora_model = lora_model.merge_and_unload()
else:
# merge weights
for layer in lora_model.base_model.model.model.layers:
if hasattr(layer.self_attn.q_proj,'merge_weights'):
layer.self_attn.q_proj.merge_weights = true
if hasattr(layer.self_attn.v_proj,'merge_weights'):
layer.self_attn.v_proj.merge_weights = true
if hasattr(layer.self_attn.k_proj,'merge_weights'):
layer.self_attn.k_proj.merge_weights = true
if hasattr(layer.self_attn.o_proj,'merge_weights'):
layer.self_attn.o_proj.merge_weights = true
if hasattr(layer.mlp.gate_proj,'merge_weights'):
layer.mlp.gate_proj.merge_weights = true
if hasattr(layer.mlp.down_proj,'merge_weights'):
layer.mlp.down_proj.merge_weights = true
if hasattr(layer.mlp.up_proj,'merge_weights'):
layer.mlp.up_proj.merge_weights = true
lora_model.train(false)
# did we do anything?
assert not torch.allclose(first_weight_old, first_weight)
lora_model_sd = lora_model.state_dict()
del lora_model, base_model
num_shards_of_models = {'7b': 1, '13b': 2}
params_of_models = {
'7b':
{
"dim": 4096,
"multiple_of": 256,
"n_heads": 32,
"n_layers": 32,
"norm_eps": 1e-06,
"vocab_size": -1,
},
'13b':
{
"dim": 5120,
"multiple_of": 256,
"n_heads": 40,
"n_layers": 40,
"norm_eps": 1e-06,
"vocab_size": -1,
},
}
params = params_of_models[model_size]
num_shards = num_shards_of_models[model_size]
n_layers = params["n_layers"]
n_heads = params["n_heads"]
dim = params["dim"]
dims_per_head = dim // n_heads
base = 10000.0
inv_freq = 1.0 / (base ** (torch.arange(0, dims_per_head, 2).float() / dims_per_head))
def permute(w):
return (
w.view(n_heads, dim // n_heads // 2, 2, dim).transpose(1, 2).reshape(dim, dim)
)
def unpermute(w):
return (
w.view(n_heads, 2, dim // n_heads // 2, dim).transpose(1, 2).reshape(dim, dim)
)
def translate_state_dict_key(k):
k = k.replace("base_model.model.", "")
if k == "model.embed_tokens.weight":
return "tok_embeddings.weight"
elif k == "model.norm.weight":
return "norm.weight"
elif k == "lm_head.weight":
return "output.weight"
elif k.startswith("model.layers."):
layer = k.split(".")[2]
if k.endswith(".self_attn.q_proj.weight"):
return f"layers.{layer}.attention.wq.weight"
elif k.endswith(".self_attn.k_proj.weight"):
return f"layers.{layer}.attention.wk.weight"
elif k.endswith(".self_attn.v_proj.weight"):
return f"layers.{layer}.attention.wv.weight"
elif k.endswith(".self_attn.o_proj.weight"):
return f"layers.{layer}.attention.wo.weight"
elif k.endswith(".mlp.gate_proj.weight"):
return f"layers.{layer}.feed_forward.w1.weight"
elif k.endswith(".mlp.down_proj.weight"):
return f"layers.{layer}.feed_forward.w2.weight"
elif k.endswith(".mlp.up_proj.weight"):
return f"layers.{layer}.feed_forward.w3.weight"
elif k.endswith(".input_layernorm.weight"):
return f"layers.{layer}.attention_norm.weight"
elif k.endswith(".post_attention_layernorm.weight"):
return f"layers.{layer}.ffn_norm.weight"
elif k.endswith("rotary_emb.inv_freq") or "lora" in k:
return none
else:
print(layer, k)
raise notimplementederror
else:
print(k)
raise notimplementederror
def save_shards(lora_model_sd, num_shards: int):
# add the no_grad context manager
with torch.no_grad():
if num_shards == 1:
new_state_dict = {}
for k, v in lora_model_sd.items():
new_k = translate_state_dict_key(k)
if new_k is not none:
if "wq" in new_k or "wk" in new_k:
new_state_dict[new_k] = unpermute(v)
else:
new_state_dict[new_k] = v
os.makedirs(output_dir, exist_ok=true)
print(f"saving shard 1 of {num_shards} into {output_dir}/consolidated.00.pth")
torch.save(new_state_dict, output_dir + "/consolidated.00.pth")
with open(output_dir + "/params.json", "w") as f:
json.dump(params, f)
else:
new_state_dicts = [dict() for _ in range(num_shards)]
for k in list(lora_model_sd.keys()):
v = lora_model_sd[k]
new_k = translate_state_dict_key(k)
if new_k is not none:
if new_k=='tok_embeddings.weight':
print(f"processing {new_k}")
assert v.size(1)%num_shards==0
splits = v.split(v.size(1)//num_shards,dim=1)
elif new_k=='output.weight':
print(f"processing {new_k}")
splits = v.split(v.size(0)//num_shards,dim=0)
elif new_k=='norm.weight':
print(f"processing {new_k}")
splits = [v] * num_shards
elif 'ffn_norm.weight' in new_k:
print(f"processing {new_k}")
splits = [v] * num_shards
elif 'attention_norm.weight' in new_k:
print(f"processing {new_k}")
splits = [v] * num_shards
elif 'w1.weight' in new_k:
print(f"processing {new_k}")
splits = v.split(v.size(0)//num_shards,dim=0)
elif 'w2.weight' in new_k:
print(f"processing {new_k}")
splits = v.split(v.size(1)//num_shards,dim=1)
elif 'w3.weight' in new_k:
print(f"processing {new_k}")
splits = v.split(v.size(0)//num_shards,dim=0)
elif 'wo.weight' in new_k:
print(f"processing {new_k}")
splits = v.split(v.size(1)//num_shards,dim=1)
elif 'wv.weight' in new_k:
print(f"processing {new_k}")
splits = v.split(v.size(0)//num_shards,dim=0)
elif "wq.weight" in new_k or "wk.weight" in new_k:
print(f"processing {new_k}")
v = unpermute(v)
splits = v.split(v.size(0)//num_shards,dim=0)
else:
print(f"unexpected key {new_k}")
raise valueerror
for sd,split in zip(new_state_dicts,splits):
sd[new_k] = split.clone()
del split
del splits
del lora_model_sd[k],v
gc.collect() # effectively enforce garbage collection
os.makedirs(output_dir, exist_ok=true)
for i,new_state_dict in enumerate(new_state_dicts):
print(f"saving shard {i+1} of {num_shards} into {output_dir}/consolidated.0{i}.pth")
torch.save(new_state_dict, output_dir + f"/consolidated.0{i}.pth")
with open(output_dir + "/params.json", "w") as f:
print(f"saving params.json into {output_dir}/params.json")
json.dump(params, f)
save_shards(lora_model_sd=lora_model_sd, num_shards=num_shards)执行命令（如果你使用了anaconda，执行命令前请先激活环境）：
python merge_llama_with_chinese_lora.py --base_model path_to_original_llama_hf_dir --lora_model chinese-alpaca-lora-7b --output_dir path_to_output_dir参数说明：
--base_model：存放hf格式的llama模型权重和配置文件的目录（前面步骤中转的hf格式）--lora_model：扩充了中文的模型目录--output_dir：指定保存全量模型权重的目录，默认为./（合并出来的目录）- （可选）
--offload_dir：对于低内存用户需要指定一个offload缓存路径
到这里就已经合并好模型了, 目录：
接下来就准备部署吧
部署模型
我们需要先下载llama.cpp进行模型的量化, 输入以下命令:
git clone https://github.com/ggerganov/llama.cpp目录如：
重点来了, 在窗口中输入以下命令进入刚刚下载的llama.cpp
cd llama.cpp如果你是跟着教程使用scoop（包管理器）安装的mingw，请使用以下命令（不是的请往后看）：
cmake . -g "mingw makefiles"
cmake --build . --config release走完以上命令后你应该能在llama.cpp的bin目录内看到以下文件：
如果你是使用的安装包的方式安装的mingw，请使用以下命令：
mkdir build
cd build
cmake ..
cmake --build . --config release走完以上命令后在build =》release =》bin目录下应该会有以下文件：
如果没有以上的文件, 那你应该是报错了, 基本上要么就是下载依赖的地方错, 要么就是编译的地方出错, 我在这里摸索了好久
接下来在llama.cpp内新建一个zh-models文件夹, 准备生成量化版本模型
接着在窗口中输入命令将上述.pth模型权重转换为ggml的fp16格式，生成文件路径为zh-models/7b/ggml-model-f16.bin
python convert-pth-to-ggml.py zh-models/7b/ 1
进一步对fp16模型进行4-bit量化，生成量化模型文件路径为zh-models/7b/ggml-model-q4_0.bin
d:\llama\llama.cpp\bin\quantize.exe ./zh-models/7b/ggml-model-f16.bin ./zh-models/7b/ggml-model-q4_0.bin 2
到这就已经量化好了, 可以进行部署看看效果了, 部署的话如果你电脑配置好的可以选择部署f16的,否则就部署q4_0的....
d:\llama\llama.cpp\bin\main.exe -m zh-models/7b/ggml-model-q4_0.bin --color -f prompts/alpaca.txt -ins -c 2048 --temp 0.2 -n 256 --repeat_penalty 1.3在提示符 > 之后输入你的prompt，cmd/ctrl+c中断输出，多行信息以\作为行尾
部署效果：
终于写完了~
参考：
👍点赞，你的认可是我创作的动力 ！
🌟收藏，你的青睐是我努力的方向！
✏️评论，你的意见是我进步的财富！
发表评论