The time it takes for newly released, powerful open-source LLMs to be implemented and made available in SambaNova is -I feel-, too slow. I still don’t see any of the newer models from Qwen or Kimi (just as examples), available to run.
SambaNova’s technology is amazing, and I understand I may not fully grasp the work required to integrate these models into the SN infrastructure soon after their release.
However, I believe there would be significant benefits to shortening this period.
Hello @slicksroboto ,
Thank you for taking the time to share your feedback with us.
We truly appreciate your enthusiasm for our technology and your interest in accessing the latest open-source Large Language Models (LLMs) on our platform.
Our team is diligently working to onboard new models to our infrastructure, with a strong commitment to providing our customers access to the highest-quality open-source models available.
We will be announcing new model additions in the near future. Please stay tuned for updates coming soon!
Appreciate you sharing! There is definitely a lot of work the team does to make sure the models are running as optimally and efficiently and we are definitely working to improve. You can definitely expect some more announcements coming soon such as OpenAI GPT-OSS model.
I’m not the person who made this thread, but the models I would love to see on sambanova are:
GLM-4.5
Baichuan-M2-32B
dots.llm1.inst
I should mention though, that the company behind GLM-4.5 (Zhipu) is under US “export blacklist” by the US department of commerce, unfortunately. So you guys might want to check with someone who can give legal advice about that first.
I have seen a few folks excited about GLM. At the moment, these are not on the roadmap, but would love to understand you are building and why these models.
I think that you have already answered my on the premise of my comments. Looking forward of seeing more (and perhaps more quickly ) newer OSLLMs when available!
@vasanth.mohan At the moment I am exploring models prior to building, but I do already know what field it will eventually be in, and that is health. I am not a doctor but I do have a great interest in this topic.
And that’s where GLM-4.5 shines for me. The last model that was great at this topic was the Llama 405B model, and that one is becoming a bit outdated (relatively).
GLM-4.5 just knows a lot of detail about health that other models miss.
If I were to ask a different model that I believe isn’t good at health (such as Qwen) the answer will almost always be in this format:
A quick confirmation of the question
Followed by a general overview of the topic
Followed by a short general answer that doesn’t mention details or show much actual knowledge
And finally a useless disclaimer about needing to see a doctor, even when the question isn’t about me
This isn’t helpful for me because it doesn’t really answer the question. It’s like a non-answer essentially.
This problem isn’t exclusive to Qwen, a lot of models do this now and it’s due to a focus on only select few benchmarks.
That’s where GLM-4.5 does significantly better. It knows the latest information about health, and a lot of information about the actual details on these topics. It’s also not afraid to talk about them.
That’s super interesting! Healthcare is definitely going to be an interesting evolving topic for these LLMs. Also one where open models shine as opposed to giving your data to OpenAI, even though I know it is one of the areas they are focusing heavily with GPT-5.
) newer OSLLMs when available!