RUN THIS LLM
Search local LLM hardware requirements
CodeLlama 70B
Meta · 70B · Code
The most capable CodeLlama. Near-GPT-4 level code generation with strong debugging skills.
VRAM Requirements
| Quantization | VRAM |
|---|---|
| Q4_K_M (smallest) | 42 GB |
| Q8_0 (balanced) | 77 GB |
| FP16 (full quality) | 140 GB |
Specifications
- Parameters: 70B
- Category: Code
- Max context: 16K tokens
- System RAM: 48 GB minimum
- HuggingFace: codellama/CodeLlama-70b-Instruct-hf
Benchmarks
- HumanEval: 53 — Code generation (pass@1)
Loading interactive analysis...