LLMs · Microsoft
Phi-3.5 MoE (42B-A6.6B)
- Parameters
- 42B (6.6B active)
- Architecture
- Mixture of experts
- Max context
- 128K tokens
- License
- mit
- Released
- 2024-08
How to run it
Quantized files · GGUF
Works in LM Studio, llama.cpp, Jan and Ollama. No sign-in needed.
huggingface.co/lmstudio-community/Phi-3.5-MoE-instruct-GGUF (opens in a new tab)
ollama run hf.co/lmstudio-community/Phi-3.5-MoE-instruct-GGUFOriginal weights · Hugging Face
Full-precision files from Microsoft, for fine-tuning, vLLM, Transformers.
huggingface.co/microsoft/Phi-3.5-MoE-instruct (opens in a new tab)
Official repo · GitHub
Reference code, inference scripts and docs from Microsoft. The weights themselves download from Hugging Face.
github.com/microsoft/PhiCookBook (opens in a new tab)
git clone https://github.com/microsoft/PhiCookBookMemory needed
Weights plus KV cache and runtime overhead. Leave a little extra free for the operating system and other apps.
| Quantization | Quality | Weights | Total at 8K context | Total at 32K context | Download |
|---|---|---|---|---|---|
Q2_K |
Low | 16.7 GB | 18.4 GB | 21.6 GB | GGUF files (opens in a new tab) |
Q3_K_M |
Low | 20.5 GB | 22.2 GB | 25.4 GB | GGUF (Q3_K_L) (opens in a new tab) |
Q4_K_M |
Good | 25.5 GB | 27.2 GB | 30.4 GB | GGUF (opens in a new tab) |
Q5_K_M |
High | 29.7 GB | 31.4 GB | 34.6 GB | GGUF files (opens in a new tab) |
Q6_K |
High | 34.3 GB | 36.0 GB | 39.2 GB | GGUF (opens in a new tab) |
Q8_0 |
Very high | 44.4 GB | 46.1 GB | 49.3 GB | GGUF (opens in a new tab) |
F16 |
Full | 83.7 GB | 85.4 GB | 88.6 GB | Original weights (opens in a new tab) |
Best hardware for Phi-3.5 MoE (42B-A6.6B)
The least expensive GPU in our catalog that runs it without offloading is the AMD Radeon RX 7900 XT (launch price $899) at Q2_K.
| Grade · points | Hardware | Quant | Memory | Speed | Fit |
|---|---|---|---|---|---|
| 78 ptsGreat (A) · 78 of 100 points: Model quality 23.9/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 14.9/15 | Apple M4 Max (64 GB) | Q4_K_M | 31.1 GB / 48.0 GB | 43.7 tok/s | Runs well |
| 77 ptsGreat (A) · 77 of 100 points: Model quality 23.9/45, Memory fit 24.0/25, Speed 15.0/15, Memory use 15.0/15 | NVIDIA GeForce RTX 5090 | Q4_K_M | 28.8 GB / 32.0 GB | 167 tok/s | Tight fit |
| 77 ptsGreat (A) · 77 of 100 points: Model quality 25.2/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 12.2/15 | NVIDIA RTX PRO 6000 Blackwell | Q8_0 | 50.9 GB / 96.0 GB | 84.8 tok/s | Runs well |
| 77 ptsGreat (A) · 77 of 100 points: Model quality 23.9/45, Memory fit 25.0/25, Speed 13.4/15, Memory use 14.9/15 | Apple M1 Max (64 GB) | Q4_K_M | 31.1 GB / 48.0 GB | 32.0 tok/s | Runs well |
| 75 ptsGreat (A) · 75 of 100 points: Model quality 21.2/45, Memory fit 24.0/25, Speed 15.0/15, Memory use 15.0/15 | AMD Radeon RX 7900 XTX | Q3_K_M | 23.4 GB / 24.0 GB | 99.4 tok/s | Tight fit |
| 75 ptsGreat (A) · 75 of 100 points: Model quality 21.2/45, Memory fit 24.0/25, Speed 15.0/15, Memory use 15.0/15 | NVIDIA GeForce RTX 3090 | Q3_K_M | 23.4 GB / 24.0 GB | 113 tok/s | Tight fit |
| 75 ptsGreat (A) · 75 of 100 points: Model quality 21.2/45, Memory fit 24.0/25, Speed 15.0/15, Memory use 15.0/15 | NVIDIA GeForce RTX 4090 | Q3_K_M | 23.4 GB / 24.0 GB | 121 tok/s | Tight fit |
| 75 ptsGreat (A) · 75 of 100 points: Model quality 23.9/45, Memory fit 25.0/25, Speed 11.1/15, Memory use 15.0/15 | Apple M4 Pro (48 GB) | Q4_K_M | 30.1 GB / 36.0 GB | 21.9 tok/s | Runs well |
| 72 ptsGreat (A) · 72 of 100 points: Model quality 24.9/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 7.6/15 | Apple M2 Ultra (192 GB) | Q6_K | 47.5 GB / 144 GB | 44.4 tok/s | Runs well |
| 72 ptsGreat (A) · 72 of 100 points: Model quality 23.9/45, Memory fit 25.0/25, Speed 15.0/15, Memory use 8.4/15 | Apple M4 Max (128 GB) | Q4_K_M | 34.9 GB / 96.0 GB | 43.7 tok/s | Runs well |
Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.