Results for “inference”

Analyze RNA velocity in single-cell RNA-seq data with scVelo, estimating cell state transitions from unspliced/spliced mRNA dynamics, inferring trajectory directions, computing latent time, and identifying driver genes.

Measures inference latency of binarized, 8-bit, and 32-bit convolutional layers on edge devices to evaluate the efficiency and speedup of the Larq Compute Engine framework compared to standard implementations.

Fine-tune and serve Physical Intelligence OpenPI models (pi0, pi0-fast, pi0.5) using JAX or PyTorch backends for robot policy inference across ALOHA, DROID, and LIBERO environments.

Guarantee valid JSON, XML, or code structure during text generation using Pydantic models for type-safe outputs, supporting local models (Transformers, vLLM, llama.cpp) and maximizing inference speed with structured generation.

Expert guidance for Groq, the LLM inference platform that provides the fastest token generation speeds available, powered by custom LPU (Language Processing Unit) hardware. Helps developers integrate Groq's API for real-time AI applications where latency matters — chatbots, code completion, and streaming responses.

Generates and analyzes proteins using ESM3 and ESM C language models, covering sequence generation, structure prediction, inverse folding, embeddings, and function conditioning with local or cloud-based Forge API inference.

Provides domain knowledge on the Causal Dynamics Architecture (CDA), an alternative AI computing architecture based on causal graphs and Hamiltonian dynamics, with references for deep dives.

Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.

Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.

Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.

Patterns for running AI models locally in browsers using WebGPU, Transformers.js, WebLLM, and ONNX Runtime. Zero API costs, full privacy. Use when "on-device AI, browser AI, WebLLM, Transformers.js, WebGPU, edge inference, offline AI, client-side ML, ONNX web, " mentioned.

State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware design. Mamba-1 (d_state=16) and Mamba-2 (d_state=128, multi-head). Models 130M-2.8B on HuggingFace.

Compress large language models using knowledge distillation from teacher to student models. Use when deploying smaller models with retained performance, transferring GPT-4 capabilities to open-source models, or reducing inference costs. Covers temperature scaling, soft targets, reverse KLD, logit distillation, and MiniLLM training strategies.

State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware design. Mamba-1 (d_state=16) and Mamba-2 (d_state=128, multi-head). Models 130M-2.8B on HuggingFace.

Compress large language models using knowledge distillation from teacher to student models. Use when deploying smaller models with retained performance, transferring GPT-4 capabilities to open-source models, or reducing inference costs. Covers temperature scaling, soft targets, reverse KLD, logit distillation, and MiniLLM training strategies.

Run read-only deep repository analysis and return a ranked synthesis with explicit confidence, concrete file references, and clear evidence-vs-inference boundaries. Use when a user says 'analyze', 'investigate', 'why does', 'what's causing', or needs grounded cross-file explanation before any changes are proposed.

Use when aer-identification has fixed the design, after methodology choice and before aer-robustness or aer-tables-figures, to run an AER-track analysis with StatsPAI — the agent-native Python engine and MCP server for causal inference, robustness, sensitivity, and publication-ready table export.

Turns a Bankr agent wallet into a keyless identity on the SIGNA agent network: resolve any identity to a messageable wallet, send and read wallet-signed DMs, invoke capabilities, and run a decentralized brain.

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA is unavailable. Supports GGUF quantization (1.5-8 bit) for reduced memory and 4-10× speedup vs PyTorch on CPU.

Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.

Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.

Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.

Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want faster inference. Supports INT8, NF4, FP4 formats, QLoRA training, and 8-bit optimizers. Works with HuggingFace Transformers.

Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory reduction with <2% perplexity degradation, or for faster inference (3-4× speedup) vs FP16. Integrates with transformers and PEFT for QLoRA fine-tuning.

Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Use when training/running transformers with long sequences (>512 tokens), encountering GPU memory issues with attention, or need faster inference. Supports PyTorch native SDPA, flash-attn library, H100 FP8, and sliding window attention.

Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Use when training/running transformers with long sequences (>512 tokens), encountering GPU memory issues with attention, or need faster inference. Supports PyTorch native SDPA, flash-attn library, H100 FP8, and sliding window attention.

Plans and coordinates the deployment of a model to Amazon SageMaker AI, selecting the appropriate pathway (real-time, serverless, async, batch, or Bedrock CMI) based on model type, traffic, latency, and cost constraints.

Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.

Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limited GPU memory, when you need faster inference than GPTQ with better accuracy preservation, or for instruction-tuned and multimodal models. MLSys 2024 Best Paper Award winner.

You are an expert in dlt, the open-source Python library for building data pipelines. You help developers load data from any API, file, or database into warehouses and lakes using simple Python decorators — with automatic schema inference, incremental loading, and built-in data contracts. dlt is the "requests library for data pipelines."

Deploy and operate the VSS 3.2 GA RT-Embed Video Embedding microservice using Docker Compose, covering GPU prerequisites, REST API usage for file uploads, text/video embeddings, live RTSP streams, Redis/Kafka/OTel integration, and troubleshooting.

Classifies data as pseudonymised or anonymised using Recital 26 reasonably likely test, Breyer ruling C-582/14, motivated intruder test, and WP29 Opinion 05/2014 on anonymisation techniques. Covers singling out, linkability, and inference tests. Keywords: pseudonymisation, anonymisation, Recital 26, re-identification, k-anonymity, differential privacy, WP29 Opinion 05/2014.

Use for writing, reviewing, debugging, testing, or optimizing Python CVXPY optimization models. Trigger on Variable, Parameter, Expression, Constraint, Objective, Problem, DCP, DPP, DGP, DQCP, solver selection/status, dual values, mixed-integer, cone, or repeated parametric solves. Do not use for scipy.optimize-only, PyMC inference, symbolic algebra without optimization, or hand-written solver implementations.

Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense models), implementing sparse architectures like Mixtral 8x7B or DeepSeek-V3, or scaling model capacity without proportional compute increase. Covers MoE architectures, routing mechanisms, load balancing, expert parallelism, and inference optimization.