Rate Limit Increase Request: Developer Research & Model Fine-Tuning

AI & Data Science NVIDIA NIM Models
1

Hello NVIDIA API Support Team,

I am an individual developer working on a personal research project involving large-scale instruction/response dataset generation for a specific domain-expert fine-tuning pipeline. I am requesting a rate limit increase to help complete a one-time generation phase for my local model development.

Project Context: I am building a high-precision domain-specific dataset (~100,000 entries) to fine-tune a local 7B parameter model. My pipeline is currently running on moonshotai/kimi-k2-thinking, which is excellent for this task but high in latency.

Technical Hurdle: Because the Kimi-K2 model has long internal “thinking” cycles, I need to run high concurrency (20 workers) to maintain a reasonable generation timeline. However, the 40 RPM ceiling on the developer tier is causing frequent 429 errors which are significantly extending the generation runtime.

Request Details:

This is a personal, non-commercial project focused on exploring the limits of specialized dataset generation. Increasing the limit would allow me to finish this specific data generation phase without 429-related interruptions.

Thank you very much for your time and for supporting individual developers on your platform.

Best regards, Joseph Saghbini.

2

To assist with this request, my NVIDIA Cloud Account ID is 0928046057701069. This is for the moonshotai/kimi-k2-thinking model on the https://integrate.api.nvidia.com/v1 endpoint.

3

Hi @TomNVIDIA and @sophwats,

Following up on my request (posted ~48 hours ago) for a temporary RPM increase to 200 RPM for the moonshotai/kimi-k2-thinking model.

I am currently hitting consistent 429 errors at the 40 RPM limit, which is stalling my research dataset generation (~100k entries).

Account Details for Review:

  • NVIDIA Cloud Account ID: 0928046057701069

  • Email: jsaghbini@outlook.com

  • Context: High-concurrency (20 workers) needed to offset the long thinking cycles of this specific model.

  • Duration: This is a one-time burst for a personal research project.

Thank you for your help in supporting individual developers on the platform!

4

Hi @Aharpster,

Can you please chime in here?

5

Hi @Aharpster, thank you for looking into this,

To clarify the request: I am an individual developer working on a personal research project (dataset generation). Because the kimi-k2-thinking model has high-latency reasoning cycles, the standard 40 RPM limit is causing constant 429 errors even with low throughput.

Account Details:

  • NVIDIA Cloud Account ID: 0928046057701069

  • Email: jsaghbini@outlook.com

  • Target: Increase from 40 RPM to 200 RPM for a one-time burst generation (~100k entries).

I have exponential backoff logic ready to go to ensure I stay within the new limit. Happy to provide any other details you need!