Please tell me whether to deploy ai in Mac studio, use ollama or lm studio

Ask Mac studio to deploy ai, is it better to use ollama or lm studio? I don't know much, but I hope my friend can give the answer to my colleague a little reason! ~

Replies

Anonymous4334 May 15, 2025

Non-technical personnel use ollama, which is simple to deploy and has commonly used models. LM is generally for technicians to toss, the number of models will be more than ollama, and there are more sources for model selection (go directly to Hugging Face to download the model)

Anonymous4335 May 15, 2025

My suggestion is lmstudio. My M4 Max uses lmstudio to deploy qwen-30b-a3b in mlx format with 95t/s, and ollama deploys qwen-30b-a3b with about 75t/s.

Anonymous472 May 15, 2025

I use ollama, which is simple and has rich models Don't do fine-tuning or the like, mainly learn MCP and Agent

Anonymous4336 May 15, 2025

mlx format takes precedence, select lm studio

Anonymous3194 May 15, 2025

I suggest API. The performance of mac is too low, the model is too small and impractical, it is too big to run, the concurrency of ram is low, and it cannot run in training. It's useless except playing all at once.

Anonymous3758 May 15, 2025

Isn't lm studio simpler? In some network isolation environments, independent deployment is very necessary.

Anonymous339 May 15, 2025

Performance is too low, no deployment value, api or web version is cheap

Anonymous4337 May 16, 2025

APIs are cheaper than electricity, unless you want to fine-tune them yourself or worry about privacy issues, there is no need to deploy them locally

Anonymous4338 May 16, 2025

Using LM Studio, with MLX and DWQ quantization (the latest quantization method, the quality is very good) is very good, and it is also convenient to open KV Cache. I use M2 Ultra to run the 4bit version of Qwen3-30B-A3B-DWQ to 105token/s, Qwen3-235B-A22B-DWQ 4bit version can also reach 32token/s

Anonymous1839 May 19, 2025

ollama saves trouble

Poster May 19, 2025

@ Anonymous4334 Good drop Thanks a lot

Poster May 19, 2025

@ Anonymous4335 tried it does exactly what you said

Poster May 19, 2025

@ Anonymous472 Good drop

Poster May 19, 2025

@ Anonymous4336 ollama doesn't have this format?

Poster May 19, 2025

@ Anonymous3194 Private

Poster May 19, 2025

@ Anonymous3758 Yes that's so right

Poster May 19, 2025

@ Anonymous4337 PrivatePrivate

Poster May 19, 2025

@ Anonymous4338 Good idea

Poster May 19, 2025

@ Anonymous1839 OK thanks a lot