Llama Cpp

orchestra-research/ai-research-skills/12-inference-serving/llama-cpp

by orchestra-research773a52944ba4MIT13K starsListed Oct 8, 2026Updated Oct 8, 2026Repository updated 3 months ago

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.

Instructions onlyAI & Agents DevOps & Cloud

AI-generated overview

Guides running local LLM inference with llama.cpp on CPUs, Apple Silicon, and non-NVIDIA GPUs.

What it does: This skill provides instructions for installing and using llama.cpp, a C/C++ LLM inference engine, including build options for Metal, CUDA, and ROCm. It covers downloading or converting GGUF models, running chat and batch inference, launching an OpenAI-compatible server, and selecting quantization formats. It also includes hardware acceleration notes, performance benchmarks, and supported model families.
When to use it: Use it when running LLM inference on CPU-only machines, Apple Silicon Macs, AMD or Intel GPUs, or edge devices without NVIDIA CUDA hardware. It is also relevant when you need GGUF quantization to reduce memory use or want a simple deployment without Docker or Python.
Requirements: Requires llama.cpp installed via Homebrew or built from source, and a GGUF model file. Optional acceleration requires Metal, CUDA, or ROCm build support. The skill lists llama-cpp-python as a dependency and references Hugging Face for model downloads. It ships no scripts; it is instructions only.

llama.cpp

Pure C/C++ LLM inference with minimal dependencies, optimized for CPUs and non-NVIDIA hardware.

When to use llama.cpp

Use llama.cpp when:

Running on CPU-only machines
Deploying on Apple Silicon (M1/M2/M3/M4)
Using AMD or Intel GPUs (no CUDA)
Edge deployment (Raspberry Pi, embedded systems)
Need simple deployment without Docker/Python

Use TensorRT-LLM instead when:

Have NVIDIA GPUs (A100/H100)
Need maximum throughput (100K+ tok/s)
Running in datacenter with CUDA

Use vLLM instead when:

Have NVIDIA GPUs
Need Python-first API
Want PagedAttention

Quick start

Installation

bash

# macOS/Linuxbrew install llama.cpp
# Or build from sourcegit clone https://github.com/ggerganov/llama.cppcd llama.cppmake
# With Metal (Apple Silicon)make LLAMA_METAL=1
# With CUDA (NVIDIA)make LLAMA_CUDA=1
# With ROCm (AMD)make LLAMA_HIP=1

# macOS/Linuxbrew install llama.cpp
# Or build from sourcegit clone https://github.com/ggerganov/llama.cppcd llama.cppmake
# With Metal (Apple Silicon)make LLAMA_METAL=1
# With CUDA (NVIDIA)make LLAMA_CUDA=1
# With ROCm (AMD)make LLAMA_HIP=1

Download model

bash

# Download from HuggingFace (GGUF format)huggingface-cli download \    TheBloke/Llama-2-7B-Chat-GGUF \    llama-2-7b-chat.Q4_K_M.gguf \    --local-dir models/
# Or convert from HuggingFacepython convert_hf_to_gguf.py models/llama-2-7b-chat/

# Download from HuggingFace (GGUF format)huggingface-cli download \    TheBloke/Llama-2-7B-Chat-GGUF \    llama-2-7b-chat.Q4_K_M.gguf \    --local-dir models/
# Or convert from HuggingFacepython convert_hf_to_gguf.py models/llama-2-7b-chat/

Run inference

bash

# Simple chat./llama-cli \    -m models/llama-2-7b-chat.Q4_K_M.gguf \    -p "Explain quantum computing" \    -n 256  # Max tokens
# Interactive chat./llama-cli \    -m models/llama-2-7b-chat.Q4_K_M.gguf \    --interactive

# Simple chat./llama-cli \    -m models/llama-2-7b-chat.Q4_K_M.gguf \    -p "Explain quantum computing" \    -n 256  # Max tokens
# Interactive chat./llama-cli \    -m models/llama-2-7b-chat.Q4_K_M.gguf \    --interactive

Server mode

bash

# Start OpenAI-compatible server./llama-server \    -m models/llama-2-7b-chat.Q4_K_M.gguf \    --host 0.0.0.0 \    --port 8080 \    -ngl 32  # Offload 32 layers to GPU
# Client requestcurl http://localhost:8080/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "llama-2-7b-chat",    "messages": [{"role": "user", "content": "Hello!"}],    "temperature": 0.7,    "max_tokens": 100  }'

# Start OpenAI-compatible server./llama-server \    -m models/llama-2-7b-chat.Q4_K_M.gguf \    --host 0.0.0.0 \    --port 8080 \    -ngl 32  # Offload 32 layers to GPU
# Client requestcurl http://localhost:8080/v1/chat/completions \  -H "Content-Type: application/json" \  -d '{    "model": "llama-2-7b-chat",    "messages": [{"role": "user", "content": "Hello!"}],    "temperature": 0.7,    "max_tokens": 100  }'

Quantization formats

GGUF format overview

Format	Bits	Size (7B)	Speed	Quality	Use Case
Q4_K_M	4.5	4.1 GB	Fast	Good	Recommended default
Q4_K_S	4.3	3.9 GB	Faster	Lower	Speed critical
Q5_K_M	5.5	4.8 GB	Medium	Better	Quality critical
Q6_K	6.5	5.5 GB	Slower	Best	Maximum quality
Q8_0	8.0	7.0 GB	Slow	Excellent	Minimal degradation
Q2_K	2.5	2.7 GB	Fastest	Poor	Testing only

Choosing quantization

bash

# General use (balanced)Q4_K_M  # 4-bit, medium quality
# Maximum speed (more degradation)Q2_K or Q3_K_M
# Maximum quality (slower)Q6_K or Q8_0
# Very large models (70B, 405B)Q3_K_M or Q4_K_S  # Lower bits to fit in memory

# General use (balanced)Q4_K_M  # 4-bit, medium quality
# Maximum speed (more degradation)Q2_K or Q3_K_M
# Maximum quality (slower)Q6_K or Q8_0
# Very large models (70B, 405B)Q3_K_M or Q4_K_S  # Lower bits to fit in memory

Hardware acceleration

Apple Silicon (Metal)

bash

# Build with Metalmake LLAMA_METAL=1
# Run with GPU acceleration (automatic)./llama-cli -m model.gguf -ngl 999  # Offload all layers
# Performance: M3 Max 40-60 tokens/sec (Llama 2-7B Q4_K_M)

# Build with Metalmake LLAMA_METAL=1
# Run with GPU acceleration (automatic)./llama-cli -m model.gguf -ngl 999  # Offload all layers
# Performance: M3 Max 40-60 tokens/sec (Llama 2-7B Q4_K_M)

NVIDIA GPUs (CUDA)

bash

# Build with CUDAmake LLAMA_CUDA=1
# Offload layers to GPU./llama-cli -m model.gguf -ngl 35  # Offload 35/40 layers
# Hybrid CPU+GPU for large models./llama-cli -m llama-70b.Q4_K_M.gguf -ngl 20  # GPU: 20 layers, CPU: rest

# Build with CUDAmake LLAMA_CUDA=1
# Offload layers to GPU./llama-cli -m model.gguf -ngl 35  # Offload 35/40 layers
# Hybrid CPU+GPU for large models./llama-cli -m llama-70b.Q4_K_M.gguf -ngl 20  # GPU: 20 layers, CPU: rest

AMD GPUs (ROCm)

Common patterns

Batch processing

bash

# Process multiple prompts from filecat prompts.txt | ./llama-cli \    -m model.gguf \    --batch-size 512 \    -n 100

# Process multiple prompts from filecat prompts.txt | ./llama-cli \    -m model.gguf \    --batch-size 512 \    -n 100

Constrained generation

bash

# JSON output with grammar./llama-cli \    -m model.gguf \    -p "Generate a person: " \    --grammar-file grammars/json.gbnf
# Outputs valid JSON only

# JSON output with grammar./llama-cli \    -m model.gguf \    -p "Generate a person: " \    --grammar-file grammars/json.gbnf
# Outputs valid JSON only

Context size

bash

# Increase context (default 512)./llama-cli \    -m model.gguf \    -c 4096  # 4K context window
# Very long context (if model supports)./llama-cli -m model.gguf -c 32768  # 32K context

# Increase context (default 512)./llama-cli \    -m model.gguf \    -c 4096  # 4K context window
# Very long context (if model supports)./llama-cli -m model.gguf -c 32768  # 32K context

Performance benchmarks

CPU performance (Llama 2-7B Q4_K_M)

GPU acceleration (Llama 2-7B Q4_K_M)

Supported models

LLaMA family:

Llama 2 (7B, 13B, 70B)
Llama 3 (8B, 70B, 405B)
Code Llama

Mistral family:

Mistral 7B
Mixtral 8x7B, 8x22B

Other:

Falcon, BLOOM, GPT-J
Phi-3, Gemma, Qwen
LLaVA (vision), Whisper (audio)

Find models: https://huggingface.co/models?library=gguf

References

Quantization Guide [blocked] - GGUF formats, conversion, quality comparison
Server Deployment [blocked] - API endpoints, Docker, monitoring
Optimization [blocked] - Performance tuning, hybrid CPU+GPU

Resources

GitHub: https://github.com/ggerganov/llama.cpp
Models: https://huggingface.co/models?library=gguf
Discord: https://discord.gg/llama-cpp

Source and attribution

Source:orchestra-research/ai-research-skillsin12-inference-serving/llama-cppat commit773a529

License: MIT

Content belongs to its original authors. SourceWeft indexes it from a public repository.

Report or request removal

More from orchestra-research/ai-research-skills

Speculative Decoding

orchestra-research

Guides accelerating LLM inference with speculative decoding, Medusa multi-head prediction, and lookahead decoding.

AI & Agents13Kupdated 3 months ago

Systems Paper Writing

orchestra-research

Guides paragraph-level structure and page budgets for writing systems conference papers such as OSDI, SOSP, ASPLOS and NSDI.

Writing & Content13Kupdated 3 months ago

Moe Training

orchestra-research

Guides training and inference of Mixture of Experts models with DeepSpeed or HuggingFace, covering routing, load balancing and expert parallelism.

AI & Agents13Kupdated 3 months ago

Presenting Conference Talks

orchestra-research

Generates conference presentation slides (Beamer LaTeX PDF and editable PPTX) from a compiled paper with speaker notes and talk script. Use when preparing oral talks, spotlight presentations, or invited talks for ML and systems conferences.

Awaiting classification13Kupdated 3 months ago

Model Pruning

orchestra-research

Guides LLM pruning with Wanda, SparseGPT, magnitude and N:M methods to cut model size and speed inference.

AI & Agents13Kupdated 3 months ago

Outlines

orchestra-research

Guides use of the Outlines library for structured, schema-constrained text generation with local models.

AI & Agents13Kupdated 3 months ago

CPU	Threads	Speed	Cost
Apple M3 Max	16	50 tok/s	$0 (local)
AMD Ryzen 9 7950X	32	35 tok/s	$0.50/hour
Intel i9-13900K	32	30 tok/s	$0.40/hour
AWS c7i.16xlarge	64	40 tok/s	$2.88/hour

GPU	Speed	vs CPU	Cost
NVIDIA RTX 4090	120 tok/s	3-4×	$0 (local)
NVIDIA A10	80 tok/s	2-3×	$1.00/hour
AMD MI250	70 tok/s	2×	$2.00/hour
Apple M3 Max (Metal)	50 tok/s	~Same	$0 (local)

Llama Cpp

llama.cpp

When to use llama.cpp

Quick start

Installation

Download model

Run inference

Server mode

Quantization formats

GGUF format overview

Choosing quantization

Hardware acceleration

Apple Silicon (Metal)

NVIDIA GPUs (CUDA)

AMD GPUs (ROCm)

Common patterns

Batch processing

Constrained generation

Context size

Performance benchmarks

CPU performance (Llama 2-7B Q4_K_M)

GPU acceleration (Llama 2-7B Q4_K_M)

Supported models

References

Resources

Source and attribution

More from orchestra-research/ai-research-skills

Speculative Decoding

Systems Paper Writing

Moe Training

Presenting Conference Talks

Model Pruning

Outlines

More in AI & Agents

Ppt Template Creator

Skill Development

Plugin Structure

Command Development

Claude Md Improver

Find Skills