add ollama-deepseek-r1-distill
This commit is contained in:
JamesFlare1212
@@ -25,7 +25,8 @@ languageName = "English"
|
||||
# whether to include Chinese/Japanese/Korean
|
||||
hasCJKLanguage = true
|
||||
# default amount of posts in each pages
|
||||
paginate = 12
|
||||
[pagination]
|
||||
pagerSize = 12
|
||||
# copyright description used only for seo schema
|
||||
copyright = ""
|
||||
# whether to use robots.txt
|
||||
|
||||
@@ -412,7 +412,7 @@ Due to the complexity of this assignment, it is best to carefully plan how to im
|
||||
|
||||
Since the flowchart is quite large, I have temporarily converted it into an image.
|
||||
|
||||
{{< image src="csci-1200-hw-2-flowchart-zh_cn.svg" caption="Flow Chart" >}}
|
||||
{{< image src="csci-1200-hw-2-flowchart-zh_cn.svg" width="100%" caption="Flow Chart" >}}
|
||||
|
||||
Mermaid source code as follows:
|
||||
|
||||
|
||||
303
content/en/posts/llama-cpp/ollama-with-deepseek-r1/index.md
Normal file
303
content/en/posts/llama-cpp/ollama-with-deepseek-r1/index.md
Normal file
@@ -0,0 +1,303 @@
|
||||
---
|
||||
title: Deploying DeepSeek R1 Distill Series Models on RTX 4090 with Ollama and Optimization
|
||||
subtitle:
|
||||
date: 2025-02-08T18:29:29-05:00
|
||||
lastmod: 2025-02-08T18:29:29-05:00
|
||||
slug: ollama-deepseek-r1-distill
|
||||
draft: false
|
||||
author:
|
||||
name: James
|
||||
link: https://www.jamesflare.com
|
||||
email:
|
||||
avatar: /site-logo.avif
|
||||
description: This blog post explores the installation, optimization, and usage of DeepSeek-R1's distilled models in Ollama on Windows 11, MacOS, and Linux, highlighting performance and limitations.
|
||||
keywords: ["DeepSeek-R1","Ollama","KV Cache","Flash Attention"]
|
||||
license:
|
||||
comment: true
|
||||
weight: 0
|
||||
tags:
|
||||
- LLM
|
||||
- llama.cpp
|
||||
- Quantization
|
||||
- Ollama
|
||||
categories:
|
||||
- LLM
|
||||
collections:
|
||||
- Ollama
|
||||
hiddenFromHomePage: false
|
||||
hiddenFromSearch: false
|
||||
hiddenFromRss: false
|
||||
hiddenFromRelated: false
|
||||
summary: This blog post explores the installation, optimization, and usage of DeepSeek-R1's distilled models in Ollama on Windows 11, MacOS, and Linux, highlighting performance and limitations.
|
||||
resources:
|
||||
- name: featured-image
|
||||
src: featured-image.jpg
|
||||
- name: featured-image-preview
|
||||
src: featured-image-preview.jpg
|
||||
toc: true
|
||||
math: false
|
||||
lightgallery: true
|
||||
password:
|
||||
message:
|
||||
repost:
|
||||
enable: false
|
||||
url:
|
||||
|
||||
# See details front matter: https://fixit.lruihao.cn/documentation/content-management/introduction/#front-matter
|
||||
---
|
||||
|
||||
<!--more-->
|
||||
|
||||
## Introduction
|
||||
|
||||
Recently, DeepSeek-R1 has gained significant attention due to its affordability and powerful performance. Additionally, the official release of several distilled models in various sizes makes it possible for consumer-grade hardware to experience the capabilities of reasoning models.
|
||||
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-7B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Llama-8B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-14B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Llama-70B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B)
|
||||
|
||||
However, it is important to note that these distilled models are far from the full DeepSeek-R1 model. For instance, `DeepSeek-R1-Distill-Qwen-32B` only reaches the level of o1-mini.
|
||||
|
||||
This can be seen in the official [chart](https://raw.githubusercontent.com/deepseek-ai/DeepSeek-R1/main/figures/benchmark.jpg) (the chart below is interactive and you can turn off data that you do not want to see).
|
||||
|
||||
{{< echarts >}}
|
||||
{
|
||||
"tooltip": {
|
||||
"trigger": "axis",
|
||||
"axisPointer": {
|
||||
"type": "shadow"
|
||||
}
|
||||
},
|
||||
"legend": {
|
||||
"top": 30,
|
||||
"data": [
|
||||
"DeepSeek-R1",
|
||||
"OpenAI-o1-1217",
|
||||
"DeepSeek-R1-32B",
|
||||
"OpenAI-o1-mini",
|
||||
"DeepSeek-V3"
|
||||
]
|
||||
},
|
||||
"grid": {
|
||||
"left": "8%",
|
||||
"right": "8%",
|
||||
"bottom": "10%",
|
||||
"containLabel": true
|
||||
},
|
||||
"xAxis": {
|
||||
"type": "category",
|
||||
"data": [
|
||||
"AIME 2024\n(Pass@1)",
|
||||
"Codeforces\n(Percentile)",
|
||||
"GPQA Diamond\n(Pass@1)",
|
||||
"MATH-500\n(Pass@1)",
|
||||
"MMLU\n(Pass@1)",
|
||||
"SWE-bench Verified\n(Resolved)"
|
||||
],
|
||||
"axisLabel": {
|
||||
"interval": 0
|
||||
}
|
||||
},
|
||||
"yAxis": {
|
||||
"type": "value",
|
||||
"min": 0,
|
||||
"max": 100,
|
||||
"name": "Accuracy / Percentile (%)",
|
||||
"nameGap": 32,
|
||||
"nameLocation": "center"
|
||||
},
|
||||
"series": [
|
||||
{
|
||||
"name": "DeepSeek-R1",
|
||||
"type": "bar",
|
||||
"data": [79.8, 96.3, 71.5, 97.3, 90.8, 49.2],
|
||||
"barGap": "0",
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "OpenAI-o1-1217",
|
||||
"type": "bar",
|
||||
"data": [79.2, 96.6, 75.7, 96.4, 91.8, 48.9],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "DeepSeek-R1-32B",
|
||||
"type": "bar",
|
||||
"data": [72.6, 90.6, 62.1, 94.3, 87.4, 36.8],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "OpenAI-o1-mini",
|
||||
"type": "bar",
|
||||
"data": [63.6, 93.4, 60.0, 90.0, 85.2, 41.6],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "DeepSeek-V3",
|
||||
"type": "bar",
|
||||
"data": [39.2, 58.7, 59.1, 90.2, 88.5, 42.0],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
{{< /echarts >}}
|
||||
|
||||
Ollama provides a convenient interface and tools for using and managing models, with the backend being llama.cpp. It supports both CPU and GPU inference optimization.
|
||||
|
||||
## Installation of Ollama
|
||||
|
||||
Follow the instructions on [Download Ollama](https://ollama.com/download) to complete the installation. My environment is as follows:
|
||||
|
||||
- Operating system: Windows 11
|
||||
- GPU: NVIDIA RTX 4090
|
||||
- CPU: Intel 13900K
|
||||
- Memory: 128G DDR5
|
||||
|
||||
## Creating Models
|
||||
|
||||
After installing Ollama, we need to create models. One way is to pull from the [Ollama Library](https://ollama.com/library/deepseek-r1:32b-qwen-distill-q4_K_M).
|
||||
|
||||
```bash
|
||||
ollama pull deepseek-r1:32b-qwen-distill-q4_K_M
|
||||
```
|
||||
|
||||
However, the default context length of this pulled model is 4096. This is insufficient and unreasonable, so we need to modify it.
|
||||
|
||||
One way is to directly edit the Modelfile. If you do not know where a model's Modelfile is located, execute the following command to view its Modelfile.
|
||||
|
||||
```bash
|
||||
ollama show --modelfile deepseek-r1:32b-qwen-distill-q4_K_M
|
||||
```
|
||||
|
||||
Here I provide my used Modelfile, which can be saved in a new text file, for example `DeepSeek-R1-Distill-Qwen-32B-Q4_K_M.txt`.
|
||||
|
||||
```text
|
||||
FROM deepseek-r1:32b-qwen-distill-q4_K_M
|
||||
|
||||
TEMPLATE """{{- if .System }}{{ .System }}{{ end }}
|
||||
{{- range $i, $_ := .Messages }}
|
||||
{{- $last := eq (len (slice $.Messages $i)) 1}}
|
||||
{{- if eq .Role "user" }}<｜User｜>{{ .Content }}
|
||||
{{- else if eq .Role "assistant" }}<｜Assistant｜>{{ .Content }}{{- if not $last }}<｜end▁of▁sentence｜>{{- end }}
|
||||
{{- end }}
|
||||
{{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{- end }}
|
||||
{{- end }}"""
|
||||
PARAMETER stop <｜begin▁of▁sentence｜>
|
||||
PARAMETER stop <｜end▁of▁sentence｜>
|
||||
PARAMETER stop <｜User｜>
|
||||
PARAMETER stop <｜Assistant｜>
|
||||
PARAMETER num_ctx 16000
|
||||
```
|
||||
|
||||
It contains several parts, and we only need to modify the `FROM` statement (indicating which model is used for construction) and the value of `num_ctx` (default 4096 unless set otherwise through API requests). Here I set it to `16000`, which represents the context length. The longer the context, the more memory and computational resources are consumed.
|
||||
|
||||
> [!NOTE]
|
||||
>
|
||||
> After testing, RTX 4090 can run a 32B q4_K_M quantized model with KV Cache quantified as q8_0 and Flash Attention enabled while maintaining a context length of 16K. If running the same configuration for a 14B q4_K_M quantized model, it can achieve a context length of 64K. I will explain more about KV Cache quantization and Flash Attention later.
|
||||
|
||||
After creating the Modelfile, we can create the model using the following command:
|
||||
|
||||
```bash
|
||||
ollama create DeepSeek-R1-Distill-Qwen-32B-Q4_K_M -f DeepSeek-R1-Distill-Qwen-32B-Q4_K_M.txt
|
||||
```
|
||||
|
||||
> [!TIP]
|
||||
>
|
||||
> The format is as follows:
|
||||
> `ollama create <name of the model to be created> -f <path and name of Modelfile>`
|
||||
|
||||
During this process, Ollama will pull the model and create it. After completion, you can execute `ollama list` to check the model list, and you should see something similar.
|
||||
|
||||
```console
|
||||
PS C:\Users\james\Desktop\Ollama> ollama list
|
||||
NAME ID SIZE MODIFIED
|
||||
DeepSeek-R1-Distill-Qwen-32B-Q4_K_M:latest ca51e8a9d628 19 GB 2 days ago
|
||||
deepseek-r1:32b-qwen-distill-q4_K_M 5de93a84837d 19 GB 2 days ago
|
||||
```
|
||||
|
||||
## Optimization
|
||||
|
||||
Ollama supports multiple optimization parameters controlled by environment variables.
|
||||
|
||||
- `OLLAMA_FLASH_ATTENTION`: Set to `1` to enable, and `0` to disable.
|
||||
- `OLLAMA_HOST`: IP address Ollama listens on. Default is `127.0.0.1`, change it to `0.0.0.0` if you want to serve externally.
|
||||
- `OLLAMA_KV_CACHE_TYPE`: Set to `q8_0` or `q4_0`. The default value is `fp16`.
|
||||
- `OLLAMA_NUM_PARALLEL`: Number of parallel requests, more throughput but higher memory consumption. Generally set to `1`.
|
||||
- `OLLAMA_ORIGINS`: CORS cross-origin request settings.
|
||||
|
||||
Flash Attention must be enabled. I recommend setting `OLLAMA_KV_CACHE_TYPE` to `q8_0`. In my tests, `q4_0` reduces the reasoning length of R1, possibly because longer content and context are more important.
|
||||
|
||||
### Windows 11
|
||||
|
||||
To set environment variables on Windows 11, go to "Advanced System Settings," then choose "Environment Variables." After that, select "New" to add a new variable. Restart Ollama for changes to take effect.
|
||||
|
||||
### MacOS
|
||||
|
||||
On MacOS, you can use commands like the following:
|
||||
|
||||
```bash
|
||||
launchctl setenv OLLAMA_FLASH_ATTENTION "1"
|
||||
launchctl setenv OLLAMA_KV_CACHE_TYPE "q8_0"
|
||||
```
|
||||
|
||||
Restart Ollama after setting environment variables.
|
||||
|
||||
### Linux
|
||||
|
||||
In Linux, modify `ollama.service` file to change its environment variables after installing Ollama:
|
||||
|
||||
```bash
|
||||
sudo systemctl edit ollama.service
|
||||
```
|
||||
|
||||
Then add the `Environment` field under `[Service]`, like this:
|
||||
|
||||
```text
|
||||
[Service]
|
||||
Environment="OLLAMA_FLASH_ATTENTION=1"
|
||||
Environment="OLLAMA_KV_CACHE_TYPE=q8_0"
|
||||
```
|
||||
|
||||
Save and reload changes:
|
||||
|
||||
```bash
|
||||
sudo systemctl daemon-reload
|
||||
sudo systemctl restart ollama
|
||||
```
|
||||
|
||||
## Limitations
|
||||
|
||||
The backend llama.cpp used by Ollama is not designed for high-concurrency and high-performance production environments. For example, its support for multi-GPU is suboptimal; it splits model layers across multiple GPUs to solve memory issues but only one GPU works at a time. To utilize the performance of multiple GPUs simultaneously, tensor parallelism is required, which SGLang or vLLM are better suited for.
|
||||
|
||||
In terms of performance, Ollama does not match SGLang or vLLM in throughput and multi-modal model support is limited with slow adaptation progress.
|
||||
|
||||
## Clients
|
||||
|
||||
For easier use of models within Ollama, I recommend two clients. Cherry Studio is a local client that I find useful, while LobeChat is a cloud-based client (I previously wrote an article on deploying the database version of LobeChat using Docker Compose).
|
||||
|
||||
{{< gh-repo-card-container >}}
|
||||
{{< gh-repo-card repo="CherryHQ/cherry-studio" >}}
|
||||
{{< gh-repo-card repo="lobehub/lobe-chat" >}}
|
||||
{{< gh-repo-card repo="Calcium-Ion/new-api" >}}
|
||||
{{< gh-repo-card repo="immersive-translate/immersive-translate" >}}
|
||||
{{< /gh-repo-card-container >}}
|
||||
|
||||
New API is a tool that I find useful for managing APIs and providing services in the OpenAI API format. Immersive Translate is another highly-rated translation plugin that supports calling OpenAI API for translations, which can also be combined with Ollama and New API. Its translation quality far exceeds traditional methods.
|
||||
@@ -406,7 +406,7 @@ A: 与 Uber 相同。保留一位小数。直接截断即可。例如，如果
|
||||
|
||||
由于流程图比较大，我暂时把它转换成了图片。
|
||||
|
||||
{{< image src="csci-1200-hw-2-flowchart-zh_cn.svg" caption="Flow Chart" >}}
|
||||
{{< image src="csci-1200-hw-2-flowchart-zh_cn.svg" width="100%" caption="Flow Chart" >}}
|
||||
|
||||
Mermaid 源码如下：
|
||||
|
||||
|
||||
@@ -22,7 +22,7 @@ tags:
|
||||
categories:
|
||||
- 大语言模型
|
||||
collections:
|
||||
- Ollama
|
||||
- LLM
|
||||
hiddenFromHomePage: false
|
||||
hiddenFromSearch: false
|
||||
hiddenFromRss: false
|
||||
|
||||
308
content/zh-cn/posts/llama-cpp/ollama-with-deepseek-r1/index.md
Normal file
308
content/zh-cn/posts/llama-cpp/ollama-with-deepseek-r1/index.md
Normal file
@@ -0,0 +1,308 @@
|
||||
---
|
||||
title: 使用 Ollama 在RTX 4090上部署 DeepSeek R1 Distill 系列模型并优化
|
||||
subtitle:
|
||||
date: 2025-02-08T18:29:29-05:00
|
||||
lastmod: 2025-02-08T18:29:29-05:00
|
||||
slug: ollama-deepseek-r1-distill
|
||||
draft: false
|
||||
author:
|
||||
name: James
|
||||
link: https://www.jamesflare.com
|
||||
email:
|
||||
avatar: /site-logo.avif
|
||||
description: 本篇文章详细介绍了如何利用DeepSeek-R1及其蒸馏模型在消费级硬件上的应用，并探讨了其性能优化和不足之处。同时提供了安装Ollama及创建深度定制化模型的步骤，以及一些提高运行效率的方法，包括使用Flash Attention和KV Cache量化等技巧。
|
||||
keywords: ["DeepSeek-R1","Ollama","KV Cache","Flash Attention"]
|
||||
license:
|
||||
comment: true
|
||||
weight: 0
|
||||
tags:
|
||||
- 大语言模型
|
||||
- llama.cpp
|
||||
- 量化
|
||||
- Ollama
|
||||
categories:
|
||||
- 大语言模型
|
||||
collections:
|
||||
- LLM
|
||||
hiddenFromHomePage: false
|
||||
hiddenFromSearch: false
|
||||
hiddenFromRss: false
|
||||
hiddenFromRelated: false
|
||||
summary: 本篇文章详细介绍了如何利用DeepSeek-R1及其蒸馏模型在消费级硬件上的应用，并探讨了其性能优化和不足之处。同时提供了安装Ollama及创建深度定制化模型的步骤，以及一些提高运行效率的方法，包括使用Flash Attention和KV Cache量化等技巧。
|
||||
resources:
|
||||
- name: featured-image
|
||||
src: featured-image.jpg
|
||||
- name: featured-image-preview
|
||||
src: featured-image-preview.jpg
|
||||
toc: true
|
||||
math: false
|
||||
lightgallery: true
|
||||
password:
|
||||
message:
|
||||
repost:
|
||||
enable: false
|
||||
url:
|
||||
|
||||
# See details front matter: https://fixit.lruihao.cn/documentation/content-management/introduction/#front-matter
|
||||
---
|
||||
|
||||
<!--more-->
|
||||
|
||||
## 前言
|
||||
|
||||
最近DeepSeek-R1爆火，原因有多种。不光价格便宜，性能强劲还开源。更难能可贵的是官方放出了几个蒸馏模型，包含各个尺寸。
|
||||
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-7B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Llama-8B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-8B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-14B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Qwen-32B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-32B)
|
||||
- [deepseek-ai/DeepSeek-R1-Distill-Llama-70B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama-70B)
|
||||
|
||||
这使得一般的消费级硬件也有机会体验Reasoning模型的魅力。不过请注意，这和真正的DeepSeek-R1相差甚远。即便是`DeepSeek-R1-Distill-Qwen-32B`也只是达到o1-mini级别的水平。
|
||||
|
||||
这一点可以参考官方给出的[图表](https://raw.githubusercontent.com/deepseek-ai/DeepSeek-R1/main/figures/benchmark.jpg)（下面这张图是可以交互的，你可以关闭你不想要的数据）。
|
||||
|
||||
{{< echarts >}}
|
||||
{
|
||||
"tooltip": {
|
||||
"trigger": "axis",
|
||||
"axisPointer": {
|
||||
"type": "shadow"
|
||||
}
|
||||
},
|
||||
"legend": {
|
||||
"top": 30,
|
||||
"data": [
|
||||
"DeepSeek-R1",
|
||||
"OpenAI-o1-1217",
|
||||
"DeepSeek-R1-32B",
|
||||
"OpenAI-o1-mini",
|
||||
"DeepSeek-V3"
|
||||
]
|
||||
},
|
||||
"grid": {
|
||||
"left": "8%",
|
||||
"right": "8%",
|
||||
"bottom": "10%",
|
||||
"containLabel": true
|
||||
},
|
||||
"xAxis": {
|
||||
"type": "category",
|
||||
"data": [
|
||||
"AIME 2024\n(Pass@1)",
|
||||
"Codeforces\n(Percentile)",
|
||||
"GPQA Diamond\n(Pass@1)",
|
||||
"MATH-500\n(Pass@1)",
|
||||
"MMLU\n(Pass@1)",
|
||||
"SWE-bench Verified\n(Resolved)"
|
||||
],
|
||||
"axisLabel": {
|
||||
"interval": 0
|
||||
}
|
||||
},
|
||||
"yAxis": {
|
||||
"type": "value",
|
||||
"min": 0,
|
||||
"max": 100,
|
||||
"name": "Accuracy / Percentile (%)",
|
||||
"nameGap": 32,
|
||||
"nameLocation": "center"
|
||||
},
|
||||
"series": [
|
||||
{
|
||||
"name": "DeepSeek-R1",
|
||||
"type": "bar",
|
||||
"data": [79.8, 96.3, 71.5, 97.3, 90.8, 49.2],
|
||||
"barGap": "0",
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "OpenAI-o1-1217",
|
||||
"type": "bar",
|
||||
"data": [79.2, 96.6, 75.7, 96.4, 91.8, 48.9],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "DeepSeek-R1-32B",
|
||||
"type": "bar",
|
||||
"data": [72.6, 90.6, 62.1, 94.3, 87.4, 36.8],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "OpenAI-o1-mini",
|
||||
"type": "bar",
|
||||
"data": [63.6, 93.4, 60.0, 90.0, 85.2, 41.6],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
},
|
||||
{
|
||||
"name": "DeepSeek-V3",
|
||||
"type": "bar",
|
||||
"data": [39.2, 58.7, 59.1, 90.2, 88.5, 42.0],
|
||||
"label": {
|
||||
"show": true,
|
||||
"position": "top"
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
{{< /echarts >}}
|
||||
|
||||
Ollama提供了更方便使用和管理模型的接口和工具，后端是llama.cpp。基于CPU推理优化的工具，也支持GPU。
|
||||
|
||||
{{< gh-repo-card-container >}}
|
||||
{{< gh-repo-card repo="ollama/ollama" >}}
|
||||
{{< gh-repo-card repo="ggerganov/llama.cpp" >}}
|
||||
{{< /gh-repo-card-container >}}
|
||||
|
||||
## 安装Ollama
|
||||
|
||||
这个根据[Download Ollama](https://ollama.com/download)的指引完成即可。我的环境如下：
|
||||
|
||||
- 操作系统是Windows 11
|
||||
- GPU是NVIDIA RTX 4090
|
||||
- CPU是Intel 13900K
|
||||
- 内存是128G DDR5
|
||||
|
||||
## 创建模型
|
||||
|
||||
在安装好Ollama后，我们就需要创建模型了。一种办法是直接从[Ollama Library](https://ollama.com/library/deepseek-r1:32b-qwen-distill-q4_K_M)拉取。
|
||||
|
||||
```bash
|
||||
ollama pull deepseek-r1:32b-qwen-distill-q4_K_M
|
||||
```
|
||||
|
||||
不过这样拉取的模型的默认上下文长度是4096。这显然不够用也不合理，所以我们要修改一下。
|
||||
|
||||
一种办法是直接修改Modelfile。如果你不知道一个模型的Modelfile可以执行以下命令查看它的Modelfile。
|
||||
|
||||
```bash
|
||||
ollama show --modelfile deepseek-r1:32b-qwen-distill-q4_K_M
|
||||
```
|
||||
|
||||
这里我给出我用的Modelfile，可以新建一个文本文件保存，比如叫做`DeepSeek-R1-Distill-Qwen-32B-Q4_K_M.txt`。
|
||||
|
||||
```text
|
||||
FROM deepseek-r1:32b-qwen-distill-q4_K_M
|
||||
|
||||
TEMPLATE """{{- if .System }}{{ .System }}{{ end }}
|
||||
{{- range $i, $_ := .Messages }}
|
||||
{{- $last := eq (len (slice $.Messages $i)) 1}}
|
||||
{{- if eq .Role "user" }}<｜User｜>{{ .Content }}
|
||||
{{- else if eq .Role "assistant" }}<｜Assistant｜>{{ .Content }}{{- if not $last }}<｜end▁of▁sentence｜>{{- end }}
|
||||
{{- end }}
|
||||
{{- if and $last (ne .Role "assistant") }}<｜Assistant｜>{{- end }}
|
||||
{{- end }}"""
|
||||
PARAMETER stop <｜begin▁of▁sentence｜>
|
||||
PARAMETER stop <｜end▁of▁sentence｜>
|
||||
PARAMETER stop <｜User｜>
|
||||
PARAMETER stop <｜Assistant｜>
|
||||
PARAMETER num_ctx 16000
|
||||
```
|
||||
|
||||
它包含多个部分，我们暂时用不着改太多，只需要注意`FROM`表明构建使用的模型（告诉Ollama用什么构建），以及`num_ctx`的值（默认4096，除非通过API请求的时候有额外设置）这里我设置的`16000`，它就是上下文长度，越长消耗的显存/内存，计算资源就越多。
|
||||
|
||||
> [!NOTE]
|
||||
>
|
||||
> 经过测试，RTX 4090差不多可以在KV Cache量化为q8_0，启用Flash Attention的情况下运行32B q4_K_M量化模型的同时，保持16K的上下文长度。如果同等情况下运行14B q4_K_M量化模型可以达到64K的上下文长度。有关KV Cache量化和Flash Attention的内容我会稍后讲解。
|
||||
|
||||
当我们创建好Modelfile后就可以使用如下命令创建模型了。
|
||||
|
||||
```bash
|
||||
ollama create DeepSeek-R1-Distill-Qwen-32B-Q4_K_M -f DeepSeek-R1-Distill-Qwen-32B-Q4_K_M.txt
|
||||
```
|
||||
|
||||
> [!TIP]
|
||||
>
|
||||
> 其格式如下：
|
||||
> `ollama create <要创建的模型名> -f <Modelfile的路径和名字>`
|
||||
|
||||
在此过程中Ollama会拉取模型并且创建它，完成后可以执行`ollama list`检查模型列表，你应该会看见类似的东西。
|
||||
|
||||
```console
|
||||
PS C:\Users\james\Desktop\Ollama> ollama list
|
||||
NAME ID SIZE MODIFIED
|
||||
DeepSeek-R1-Distill-Qwen-32B-Q4_K_M:latest ca51e8a9d628 19 GB 2 days ago
|
||||
deepseek-r1:32b-qwen-distill-q4_K_M 5de93a84837d 19 GB 2 days ago
|
||||
```
|
||||
|
||||
## 优化
|
||||
|
||||
Ollama支持多个优化参数，它们通过环境变量控制。
|
||||
|
||||
- `OLLAMA_FLASH_ATTENTION`：`1`开启，`0`关闭
|
||||
- `OLLAMA_HOST`：Ollama监听的IP，默认是`127.0.0.1`，如果要对外服务需要改成`0.0.0.0`
|
||||
- `OLLAMA_KV_CACHE_TYPE`：默认`fp16`，可以设置`q8_0`，或者`q4_0`
|
||||
- `OLLAMA_NUM_PARALLEL`：同时运行的请求数，越多吞吐量越大，显存/内存消耗越多，一般`1`就差不多了
|
||||
- `OLLAMA_ORIGINS`：有关CORS跨站请求的内容，如果你要在其它地方请求Ollama，特别域名不一样的话你要设置对应的域，或者设置`*`允许所有来源
|
||||
|
||||
Flash Attention是必开的，KV Cache我建议选`q8_0`，实测发现`q4_0`会让R1的思考长度下降，这可能是因为内容都比较长，上下文比较重要。
|
||||
|
||||
### Windows 11
|
||||
|
||||
要在Windows 11中设置环境变量，需要进入“高级系统设置”，然后选择“环境变量”，之后选择“新建”。重启Ollama使其生效。
|
||||
|
||||
### MacOS
|
||||
|
||||
在MacOS中可以执行诸如
|
||||
|
||||
```bash
|
||||
launchctl setenv OLLAMA_FLASH_ATTENTION "1"
|
||||
launchctl setenv OLLAMA_KV_CACHE_TYPE "q8_0"
|
||||
```
|
||||
|
||||
的命令来设置环境变量。重启Ollama使其生效。
|
||||
|
||||
### Linux
|
||||
|
||||
在Linux中，在安装完Ollama后可以修改`ollama.service`文件来修改它的环境变量。
|
||||
|
||||
```bash
|
||||
sudo systemctl edit ollama.service
|
||||
```
|
||||
|
||||
然后在`[Service]`下添加`Environment`字段，类似这样
|
||||
|
||||
```text
|
||||
[Service]
|
||||
Environment="OLLAMA_FLASH_ATTENTION=1"
|
||||
Environment="OLLAMA_KV_CACHE_TYPE=q8_0"
|
||||
```
|
||||
|
||||
保存修改后重载
|
||||
|
||||
```bash
|
||||
sudo systemctl daemon-reload
|
||||
sudo systemctl restart ollama
|
||||
```
|
||||
|
||||
## 不足
|
||||
|
||||
Ollama使用的后端llama.cpp并非是为了多并发和高性能的生产环境设计的。比如它对多GPU的支持就不是很理想，它会把模型的层拆分到多个GPU里，这样解决了显存不足的问题，但是这样导致在单一时间内，只有一块GPU在干活。要同时利用多张GPU的性能，我们需要张量并行，这是SGLang或者vLLM擅长的。
|
||||
|
||||
至于性能，在和SGLang或者vLLM对比的时候也不占优势，吞吐量远不及后者。其次对多模态模型的支持有限，适配进度缓慢。
|
||||
|
||||
## 客户端
|
||||
|
||||
为了更方便使用Ollama中的模型，我推荐两个客户端。Cherry Studio是我觉得好用的本地客户端，LobeChat是我觉得好用的云端客户端（我之前写过一篇 [使用 Docker Compose 部署 LobeChat 服务端数据库版本](../install-lobechat-db/)）
|
||||
|
||||
{{< gh-repo-card-container >}}
|
||||
{{< gh-repo-card repo="CherryHQ/cherry-studio" >}}
|
||||
{{< gh-repo-card repo="lobehub/lobe-chat" >}}
|
||||
{{< gh-repo-card repo="Calcium-Ion/new-api" >}}
|
||||
{{< gh-repo-card repo="immersive-translate/immersive-translate" >}}
|
||||
{{< /gh-repo-card-container >}}
|
||||
|
||||
New API则是我觉得一个很好的，用来集中管理API并且以OpenAI API格式提供服务的工具。Immersive Translate则是一个好评如潮的翻译插件，它支持调用OpenAI API来进行翻译，也自然可以与Ollama以及New API组合搭配。翻译效果远超传统翻译方法。
|
||||
Reference in New Issue
Block a user