Are there plans for a ~30B or 32B parameter version of Intern-S1?
- Dominant language
- No language data
- Stars
- 861
- Forks
- 51
- Avg merge
- 1m
- Merged PRs (30d)
- 1
Description
## Background & Positive Feedback
First of all, thank you for the amazing work on Intern-S1! We have been testing the Intern-S1-Mini version, and the performance is quite impressive, especially its reasoning capabilities at such a compact scale.
## The Request
Given the success of the Mini version, we are planning to further fine-tune and deploy Intern-S1 for our domain-specific tasks. However, we are facing a common challenge:
The smaller models (like Mini) sometimes lack the necessary parameter density for deep reasoning in complex vertical domains.
The larger versions are too resource-intensive for both fine-tuning and on-premise deployment, requiring extensive multi-GPU clusters that are not always accessible.
Therefore, we are wondering if the team has plans to release a 30B or 32B version of Intern-S1?
## Why 32B?
We believe a 32B model would be the "sweet spot" for our research and industrial application because:
Fine-tuning efficiency: It allows for full-parameter or LoRA fine-tuning on consumer-grade or mid-range server GPUs.
Deployment: It fits perfectly into a single or dual-GPU setup after 4-bit/8-bit quantization while maintaining high reasoning quality.
We are really looking forward to a mid-sized model in the Intern-S1 series to bridge this gap. Thank you again for your contribution to the community!
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.