OllamaでQwen3をサクッと試そうとしてつまずく
Qwen3が出たというのを聞いて試してみました。
どうもすごいということらしいのでワクワク。
ollamaで使えるモデルを探す
Ollamaでモデルを使うにはOllamaで使えるGGUFのモデルを探す必要があります。
と思ったら、公式からもう出てました。仕事はやっ。
ということでこれを入れていきます。
Ollamaで動かす
今回は VRAM 192GB の Mac Studio で動かしたのでお構なく一番強いモデルを選びました。
一番強いモデルはqwen3:235b-a22bですね。早速動かしましょう。
ollama run qwen3:235b-a22b
で、容量が142GBもあることに気がついて一旦中断。
Tmuxで動かしながら別の作業をしましょう。
で、いつの間にかダウンロードが終わってTmuxウィンドウが閉じてました。
満を持して動かします。
user@115 ~ % ollama run qwen3:235b-a22b
Error: unable to load model: /Users/user/.ollama/models/blobs/sha256-aeacdadecbed8a07e42026d1a1d3cd30715bb2994ebe4e4ca4009e1a4abe8d5d
あれ?動かない。
なんだこれ?初めてみたぞ。もしかしてダウンロードがうまく行ってなかったのかなと思って確認したけど、そんなこともなかった。
user@115 ~ % ollama list
NAME ID SIZE MODIFIED
qwen3:235b-a22b 8471b4f16b88 142 GB 42 minutes ago
gemma3:27b a418f5838eaf 17 GB 2 weeks ago
minicpm-v:latest c92bfad01205 5.5 GB 6 weeks ago
llava:latest 8dd30f6b0cb1 4.7 GB 6 weeks ago
llava-phi3:latest c7edd7b87593 2.9 GB 6 weeks ago
mxbai-embed-large:latest 468836162de7 669 MB 2 months ago
kun432/cl-nagoya-ruri-large:latest 4df32234386c 675 MB 2 months ago
hf.co/mmnga/cyberagent-DeepSeek-R1-Distill-Qwen-32B-Japanese-gguf:IQ1_M 24ab4d3f7d51 7.9 GB 2 months ago
ちゃんとある。
で、いろいろ調べてみる(Error: unable to load model:)と似たようなことで詰まってる人がいました。
なるほどOllamaのバージョンが古いと。
確認してみると確かに古い。
user@115 ~ % ollama -v
ollama version is 0.6.5
Updateする。
brew upgrade ollama
で。リスタート
brew services restart ollama
OK
user@115 ~ % ollama -v
ollama version is 0.6.6
で、動かしてみる。ようやく。
ollama run qwen3:235b-a22b
動いた〜!!
結果と感想
user@115 ~ % ollama run qwen3:235b-a22b
>>> hello. who are you?
<think>
Okay, the user asked, "hello. who are you?" I need
to respond appropriately. Let me start by
understanding what they're asking. They probably
want to know what kind of AI I am. I should keep
it friendly and concise.
First, greet them back. Then introduce myself as
Qwen, the large language model developed by
Alibaba Cloud's Tongyi Lab. Mention my
capabilities like answering questions, creating
text, and various other tasks. Also, offer
assistance so they know I'm here to help. Keep the
tone approachable and not too technical. Make sure
there's an invitation for them to ask more if
needed. Check for clarity and make sure it flows
naturally. Avoid any jargon. Alright, that should
cover it.
</think>
Hello! I'm **Qwen**, a large language model
developed by Alibaba Cloud's Tongyi Lab. I can
help answer questions, create text (like stories,
emails, scripts), perform logical reasoning, code,
and more. Think of me as a versatile AI assistant
ready to help with a wide range of tasks! How can
I assist you today?
なるほど。確かに言われているようにすごく試行型のモデルっぽい。今のところローカルで動くLLMの中だと飛び抜けて文章量が多い気がする。
試しに情報の正確性を見てみる。
>>> 電気通信大学について教えて
<think>
Okay, the user asked "電気通信大学について教えて"
which means "Tell me about the University of
Electro-Communications." Let me start by recalling
what I know about this university.
First, I need to confirm the basic information.
The University of Electro-Communications, or UEC,
is located in Tokyo, Japan. It's a national
university, so it's public. Founded in 1949, but
its roots go back to the Radio Technical Training
Institute established in 1918. That's important
for the history part.
Next, academic structure. They have faculties like
Engineering, Informatics, and maybe others. Their
graduate schools would include similar
departments. The university is known for its focus
on science and technology, especially in
information technology, electronics, and
communications. They probably have a strong
research emphasis.
Campus location: Nerima City, Tokyo. I should
mention that it's in the western part of Tokyo.
Student population, both undergraduate and
graduate. Maybe around 4,000 undergrads and 1,500
grad students. International students might be a
smaller percentage, but UEC does have exchange
programs.
Research areas: They are known for information
science, AI, robotics, electronics, and maybe
cybersecurity. They have research institutes and
collaborations with industry. Alumni and notable
achievements: producing experts in tech fields,
contributions to IT and telecommunications in
Japan.
Strengths: Strong ties with the tech industry,
practical education, high employment rate. Maybe
mention partnerships with companies like Hitachi,
Fujitsu, Toyota, etc.
I should structure the answer into sections:
History, Academic Programs, Research, Campus,
International Collaboration, and Strengths. Keep
it clear and concise. Avoid overly technical terms
but include key details. Check if there's any
recent info or updates, but since my knowledge is
up to 2023, I should be fine. Also, make sure to
mention the nickname "UEC Tokyo" sometimes used.
Ensure the response flows naturally and covers all
the main points without being too verbose.
</think>
以下は、電気通信大学(UEC Tokyo)についての基本情報
です:
### :再生ボタン:歴史
- **設立**:1949年に現在の形で開学(前身は1918年創設
の「無線電信講習所」)。
- **特徴**:日本の通信・情報技術の発展に貢献し、技術
者育成の伝統を持つ国立大学。
### :再生ボタン:学部・大学院
- **学部**:
- 情報工学部(情報通信工学科、コンピュータ理工学科
など)
- 情報理工学部(機械制御工学科、電子情報物理工学科
など)
- システム工学部(社会基盤情報工学科、マネジメント
工学科など)
- 情報マネジメント学部(経営システム工学科、知的財
産学科など)
- **大学院**:
- 情報理工学研究科(博士前期・後期課程)
- システム理工学研究科
- 情報マネジメント研究科
### :再生ボタン:研究・強み
- **重点分野**:AI、ロボティクス、通信技術、量子情報
、次世代半導体、セキュリティなど。
- **研究機関**:先端メディアサイエンス研究所、情報通
信工学研究所など。
- **産業連携**:NTT、KDDI、ソニー、トヨタなど大手企
業と共同研究多数。
### :再生ボタン:キャンパス
- **所在地**:東京都調布市(多摩地域)。自然に恵まれ
た環境で、東京都心へのアクセスも便利。
- **学生数**:約5,500人(学部生約4,000人、大学院生約
1,500人)。
### :再生ボタン:国際性
- 約300人の留学生が在籍し、海外大学との交換留学プロ
グラムも充実。
- 英語での授業(Global Courseなど)も提供。
### :再生ボタン:卒業生の進路
- 情報通信業界(楽天、ドコモ、KDDIなど)、IT企業(
Google、サイバーエージェントなど)、製造業(トヨタ、
日立、三菱重工など)への就職率が高い。
### :再生ボタン:その他
- **就職実績**:情報系国立大学の中でもトップクラスの
就職率(毎年95%以上)。
- **教育の特色**:少人数教育と実験・演習科目の充実で
、「実学」を重視。
---
何か特定の分野(研究内容、入試、学生生活など)につい
て詳しく知りたい場合は、お気軽にご質問ください!
>>> Send a message (/? for help)
少し情報が古い。ここはやっぱりGPT4.1がダントツで精度が良いなと感じる。
とはいえ試行量がとんでも無く多いイメージなので確かにmcp用途には向いてそうだ。
ちょうどmcpでサービスを作っているのでmcpで使った感想もどこかで書こう。
とりあえず。使ってみました。
出た。試した。よかった。?
Discussion