What Models Would You Like to See on SambaNova Cloud?

SambaNova Devs
1

We’re excited to hear from you about the models you’d like to see available on the SambaNova Cloud. As we continue to enhance our offerings, your feedback is invaluable in shaping the future of our platform. What models are you interested in?

Please share your thoughts and suggestions in this thread. Your input will help us prioritize which models to integrate next, ensuring we meet the needs of our diverse user base. Looking forward to your responses!

6 Likes
2

I personally think openai’s GPT 4o realtime+ audio models would be amazing. imagine using realtime api when the model is running on sambanova cloud. That would allow creating lots of application in this real time translation and even simultaneous interpretation area, cause the STT-LLM-TTS architecture is not quite capable of doing real time simultaneous interpretation.

4 Likes
3

Qwen 2.5 72b and Qwen 2.5 coder 32b pls!
Deepseek v2.5 is also very good even after half a year from release, but it’s a bit more tricky to deploy

5 Likes
4

Good to hear that feedback and will make sure we get that over to the teams!

5

For me I already have some recommendations such as:

  • Qwen 2.5 coder and 72b instruct
  • 01.ai - Yi-34b
  • Tencent Hunyuan-A52B-Instruct
  • it would be also great to see an advanced embed model.

Thanks in advance!

2 Likes
6

I would add the Qwen 2.5 turbo with the 1m context length, which helps with large data processing.

1 Like
7

Based on today’s news, I would add the Phi 4.

8

Maybe just Uncensored model? I’m happy to see that

2 Likes
9

@markpaine.it Welcome to the community . Providing a completely uncensored model is not on our roadmap and I doubt if it ever will be . Without being too graphic ( I have my own filters here ) , why would you require such a thing ?

-Coby

1 Like
10

Deepseek V3, is published, 650B MoE model

7 Likes
11

can non-transformer architecture also available on sambanova cloud? I want jamba 1.5 large from ai21 lab

1 Like
12

Some text embedding model would be nice, like e5, e5mistral, etc. That would help with many RAG use cases.

3 Likes
13

The Deepseek V3 650B is absolutely the one I would most like to see on SambaNova Cloud.

3 Likes
14

definitely deepseek v3

2 Likes
15

jamba 1.5 large from ai21 lab

1 Like
16

@hxy9243 One of these is on the very near term roadmap

-Coby

1 Like
18

Must to be Deepseek V3

1 Like
19

Deepseek Coder is better for autocomplete

1 Like
20

@ricardotenv Have you used the Qwen2.5-Coder for code completion as well ? I use it in house and it is really good