need a setting that automatically hides thinking token in the api.
please give me access to deepseek r1…4k context length is too short especially for a reasoning model. for complex prompts it will enough only for a single message we require a minium context length of 8k, but 16k is the prefered.
also i would like to get the free credits promised for joining and actively participating in the community.
Great news! Thanks, Coby!
The credits are now reflecting on your account the original signup credit was already there and now the community credit. We did not state when the community credit would be applied and the batch was planned for this weekend. I went ahead and applied yours manually , early.
As to access to the DeepSeek R1 via API you need to sign up for the waitlist which can be done at the following URL which is also accessible by clicking the waitlist link in the playground where this is stated.
-Coby
Already joined the waitlist
Urgent : LinkedIn Collaboration Opportunity 120K+ Followers - LinkedIn Top Voice '24
Subject: LinkedIn Collaboration for DeepSeek API – Showcasing the Fastest AI Model
Dear Team,
I hope you’re doing well.
My name is Vikram Gaur, and I am honored to be a LinkedIn Top Voice 2024, with a thriving community of 110K+ engaged tech enthusiasts. Recently, I had the privilege of being featured in Times Square, NYC, and have successfully collaborated with leading brands across AI, cloud technologies, and startups including— Technology & AI: NVIDIA, AMD, Google Cloud, Amazon, DBS Bank, Cashfree Payments, Pepsi, Lenskart.com, Swiggy, Pieces for Developers
Currently, I work as an SDE at EY and serve as a Google Cloud Facilitator, keeping me deeply connected with the tech, AI, and developer ecosystems.
Collaboration Opportunity
I’d love to explore the possibility of collaborating with SambaNova Cloud to craft a high-impact LinkedIn post that not only highlights DeepSeek API’s groundbreaking capabilities but also resonates with my audience. My experience in curating engaging, insightful content can help amplify DeepSeek’s visibility and drive meaningful conversations in the AI and developer space.
DeepSeek-R1: The Fastest AI Model
SambaNova has launched DeepSeek-R1 671B, the fastest and most efficient model available, achieving 198 tokens per second—3X faster and 5X more efficient than the latest GPUs. This is a major breakthrough for real-time AI inference at scale, solving one of the biggest challenges in AI adoption.
DeepSeek-R1, built on a Mixture of Experts (MoE) architecture with 671 billion parameters, is revolutionizing reasoning models, providing unmatched accuracy and decision-making capabilities. With SambaNova’s proprietary dataflow architecture, inference costs have been drastically reduced, collapsing 40 racks (320 GPUs) to a single rack (16 RDUs).
Industry leaders like Dr. Andrew Ng and Blackbox AI’s CEO Robert Rizk have highlighted its game-changing impact for AI developers and enterprises. The DeepSeek API on SambaNova Cloud is now available for early access, enabling developers to experience the full model’s power.
More details: cloud.sambanova.ai
Target Audience
My LinkedIn audience includes a diverse group of professionals:
1. AI & Machine Learning Developers
- Engineers and researchers working on AI/ML models
- Developers building AI-driven applications in reasoning, search, and coding automation
- Open-source contributors and AI enthusiasts
2. Enterprises & Fortune 500 Companies
- Large organizations seeking on-premise AI deployment for privacy and security
- Enterprises using AI for automation, search, and data analytics
- AI-driven businesses optimizing cost-efficient inference solutions
3. Startups & AI-First Companies
- AI startups requiring high-performance reasoning models
- Companies in AI-powered coding, research, and content generation
- Emerging businesses building applications with DeepSeek-R1
4. Cloud & Infrastructure Teams
- Cloud architects and infrastructure engineers optimizing AI workloads
- CTOs and IT teams seeking alternatives to traditional GPUs
5. AI & Research Institutions
- Universities, AI research labs, and institutions using DeepSeek-R1 for academic research
- Researchers working on Mixture of Experts (MoE) models and AI efficiency
6. AI Product Leaders & Data Scientists
- AI product managers implementing LLM-powered solutions
- Data scientists optimizing AI models for efficiency and scalability
7. Governments & Public Sector
- Organizations focused on AI sovereignty and secure on-premise AI solutions
- Defense and research institutions requiring high-performance AI inference
They are positioning DeepSeek-R1 671B and SambaNova’s RDU chips as the best alternative to GPUs for reasoning AI models, appealing to organizations looking for faster, cost-effective, and scalable AI inference solutions.
This audience values cutting-edge AI innovations, and a well-crafted collaboration could position DeepSeek API as a powerful tool in their tech stack.
I’d love to discuss how we can work together to showcase DeepSeek API to a broader, highly engaged tech and AI community. Let’s create something impactful!
Looking forward to your thoughts.
Thank you and best regards,
Vikram Gaur
LinkedIn Top Voice '24 | Times Square Feature | India’s Top 200 LinkedIn Creator
SDE at EY | Google Cloud Facilitator | Topmate Rising Star Award Winner
gaur.vikram0023@gmail.com |
LinkedIn: Vikram Gaur - Software Development Engineer - EY | LinkedIn
That’s a great first step!
Will you be able to make the context size acceptable? Currently with only 4k it is not very usable.
Hi, I wanted to make an impassioned plea for API access.
I’m currently studying for the American Board of Surgery exam and Deepseek is helping a great deal. There are other providers for Deepseek, but nothing comes close to your speed.
Once the smoke clears from exam studying, I’m hoping to roll my own ambient clinical experience app using SambaNova. There are a lot of providers/companies in this space, and there all are equally doing a pretty crappy job, even Microsoft/Nuance (no surprise for Nuance, they had an early advantage in dictation, and absolutely sat on it for 10 years, and have been completely steamrolled by Whisper/AI).
Please let me know as soon as possible.