---
license: mit
language:
- en
base_model:
- unsloth/phi-4
- microsoft/phi-4
pipeline_tag: text-generation
---

# Phi-4 converted for ExLlamaV2 

[ExLlamaV2 is an inference library for running local LLMs on modern consumer GPUs.](https://github.com/turboderp-org/exllamav2)

| Filename | Quant type | File Size | Vram*|
| -------- | ---------- | --------- | -------- |
| [phi-4_hb8_3bpw](https://huggingface.co/cmh/phi-4_exl2/tree/hb8_3bpw) | 3.00 bits per weight | 6.66 GB | **10,3 GB** |
| [phi-4_hb8_4bpw](https://huggingface.co/cmh/phi-4_exl2/tree/hb8_4bpw) | 4.00 bits per weight | 8.36 GB | **11,9 GB** |
| [phi-4_hb8_5bpw](https://huggingface.co/cmh/phi-4_exl2/tree/hb8_5bpw) | 5.00 bits per weight | 10.1 GB | **13,5 GB** |
| [phi-4_hb8_6bpw](https://huggingface.co/cmh/phi-4_exl2/tree/hb8_6bpw) | 6.00 bits per weight | 11.8 GB | **15,1 GB** |
| [phi-4_hb8_7bpw](https://huggingface.co/cmh/phi-4_exl2/tree/hb8_7bpw) | 7.00 bits per weight | 13.5 GB | **16,7 GB** |
| [phi-4_hb8_8bpw](https://huggingface.co/cmh/phi-4_exl2/tree/hb8_8bpw) | 8.00 bits per weight | 15.2 GB | **18,2 GB** |

*at 16k context
  
# Phi-4 Model Card 

[Phi-4 Technical Report](https://arxiv.org/pdf/2412.08905)

## Model Summary 

|                         |                                                                               |     
|-------------------------|-------------------------------------------------------------------------------|
| **Developers**          | Microsoft Research                                                            |
| **Description**         | `phi-4` is a state-of-the-art open model built upon a blend of synthetic datasets, data from filtered public domain websites, and acquired academic books and Q&A datasets. The goal of this approach was to ensure that small capable models were trained with data focused on high quality and advanced reasoning.<br><br>`phi-4` underwent a rigorous enhancement and alignment process, incorporating both supervised fine-tuning and direct preference optimization to ensure precise instruction adherence and robust safety measures                |
| **Architecture**        | 14B parameters, dense decoder-only Transformer model                          |
| **Context length**      | 16384 tokens                                                                    |

## Usage

### Input Formats

Given the nature of the training data, `phi-4` is best suited for prompts using the chat format as follows: 

```bash
<|im_start|>system<|im_sep|>
You are a medieval knight and must provide explanations to modern people.<|im_end|>
<|im_start|>user<|im_sep|>
How should I explain the Internet?<|im_end|>
<|im_start|>assistant<|im_sep|>
```

### With ExUI:

Edit exui/backend/prompts.py

```
class PromptFormat_phi4(PromptFormat):

    description = "Phi-4 format"

    def __init__(self):
        super().__init__()
        pass

    def is_instruct(self):
        return True

    def stop_conditions(self, tokenizer, settings):
        return \
            [tokenizer.eos_token_id,
             """<|im_end|>"""]

    def format(self, prompt, response, system_prompt, settings):
        text = ""
        if system_prompt and system_prompt.strip() != "":
            text += "<|im_start|>system\n"
            text += system_prompt
            text += "\n<|im_end|>\n"
        text += "<|im_start|>user\n"
        text += prompt
        text += "<|im_end|>\n"
        text += "<|im_start|>assistant\n"
        if response:
            text += response
            text += "<|im_end|>\n"
        return text

    def context_bos(self):
        return True

prompt_formats = \
{
    "Chat-RP": PromptFormat_raw,
    "Llama-chat": PromptFormat_llama,
    "Llama3-instruct": PromptFormat_llama3,
    "ChatML": PromptFormat_chatml,
    "TinyLlama-chat": PromptFormat_tinyllama,
    "MistralLite": PromptFormat_mistrallite,
    "Phind-CodeLlama": PromptFormat_phind_codellama,
    "Deepseek-chat": PromptFormat_deepseek_chat,
    "Deepseek-instruct": PromptFormat_deepseek_instruct,
    "OpenChat": PromptFormat_openchat,
    "Gemma": PromptFormat_gemma,
    "Cohere": PromptFormat_cohere,
    "Phi3-instruct": PromptFormat_phi3,
    "Phi4": PromptFormat_phi4,
    "Granite": PromptFormat_granite,
    "Mistral V1": PromptFormat_mistralv1,
    "Mistral V2/V3": PromptFormat_mistralv2v3,
    "Mistral V3 (Tekken)": PromptFormat_mistralTekken,
}
```