Fireworks - Blog

DeepSeek v3 and R1 Model Architecture: Why it's powerful and economical

This article introduces the architecture of DeepSeek v3 and DeepSeek R1, which shares exactly the same architecture. Explains DeepSeek MoE (Mixture of Experts) and FP8 pre-training in depth.Fireworks boosted DeepSeek v3 and R1 inference efficiency with Multi-token prediction.

Featured Blogs

DeepSeek v3 and R1 Model Architecture: Why it's powerful and economical

DeepSeek R1 Just Got Eyes with Fireworks AI Document Inlining

From text to task: Constrained generation for structured extraction in R1

Distillation with Reasoning: Can DeepSeek R1 Teach Better Than Humans?

Mistral Small 3 Now Available on Fireworks: Faster, Lighter, and More Efficient

Beyond Supervised Fine Tuning: How Reinforcement Learning Empowers AI with Minimal Labels

DeepSeek R1: All you need to know 🐳

Fireworks Streaming Transcription: 300ms with Whisper-v3-large-quality

Real-time, performant code assistance: How Sourcegraph scaled with Fireworks AI

Document inlining: Crossing the modality gap with Compound AI

DeepSeek V3 just got vision capabilities!

20x faster Whisper than OpenAI - Fireworks audio transcribes 1 hour in 4 seconds

How Cresta drives millions of real-time, AI-powered contact center interactions with Fireworks

Fireworks f1: A breakthrough in complex reasoning with Compound AI

How Upwork and Fireworks deliver faster, smarter proposals for freelancers

FLUX.1 on Fireworks: Fast, frugal, and flexible

FireAttention V3: Enabling AMD as a viable alternative for GPU inference

Three projects, one platform: A developer's winning streak with Fireworks AI

Partnering with Meta: Bringing Llama 3.2 to Fireworks for Fine-Tuning and Inference

How Enterprises are using Multimodal Models in production with Fireworks

Multi-LoRA: Personalize AI at scale and deliver the best experience for each customer and use case, with 100x cost-efficiency

FireOptimizer: Customizing latency and quality for your production inference workload

Build Your Own Flight Recommendation System using FastAPI, SerpAPI, and Firefunction

Building a RAG with Astro, FastAPI, SurrealDB and Llama 3.1

How Fireworks evaluates quantization precisely and interpretably

Introducing Llama 3.1 inference endpoints in partnership with Meta

Fireworks AI Raises $52M Series B to Lead Industry Shift to Compound AI Systems

How Cursor built Fast Apply using the Speculative Decoding API

FireAttention V2: 12x faster to make Long Contexts practical for Online Inference

Firefunction-v2: Function calling capability on par with GPT4o at 2.5x the speed and 10% of the cost

Announcing custom models and on-demand H100s with 50%+ lower costs and latency than vLLM

GPUs on-demand: Not serverless, not reserved, but some third thing

Code Generation with Large Language Models - Fireworks AI Take

Doomed to Code: How we Teamed Up with Fireworks AI at MistralAI Hackathon to Conquer the Shores of Hell

Partnering with Meta to bring Llama 3 to Firework’s inference and fine-tuning

Getting Started with Stability’s API Powered by Fireworks

Optimizing Retrieval Augmented Generation (RAG) with MongoDB Atlas and Fireworks AI

Fireworks launches fine-tuning service - Rapidly iterate on quality and scale to production through Fireworks inference

Fireworks Platform Spring 2024 Updates

FireFunction V1 - Fireworks’ GPT-4-level function calling model - 4x faster than GPT-4 and open weights

Why do all LLMs need structured output modes?

FireLLaVA: the first commercially permissive OSS LLaVA model

FireAttention — Serving Open Source Models 4x faster than vLLM by quantizing with ~no tradeoffs

Fireworks Raises the Quality Bar with Function Calling Model and API Release

Mixtral 8x7B on Fireworks: faster, cheaper, even before the official release

LLM Inference Performance Benchmarking (Part 1)

New in Fireworks: Image-to-Image and ControlNet support for SSD-1B and SDXL!

Fireworks.ai Achieves SOC 2 Type II and HIPAA Compliance

Accelerating Code Completion with Fireworks Fast LLM Inference

Fireworks.ai Now Available on LangChain Prompt Playground

Simplifying Code Infilling with Code Llama and Fireworks.ai

Speed, Python: Pick Two. How CUDA Graphs Enable Fast Python Code for Deep Learning

Fireworks.ai: Fast, Affordable, Customizable Gen AI Platform

Multi-Query Attention is All You Need