Local inference

How to use G4F to run language models locally

Required dependencies

make sure to install the required dependencies by running

pip install g4f[local]

or

pip install -U gpt4all

---

Basic usage example

from g4f.local import LocalClient

client   = LocalClient()
response = client.chat.completions.create(
    model    = 'orca-mini-3b',
    messages = [{"role": "user", "content": "hi"}],
    stream   = True
)

for token in response:
    print(token.choices[0].delta.content or "")

Upon first use, there will be a prompt asking you if you wish to download the model. If you respond with y, G4F will go ahead and download the model for you.

You can also manually place supported models into ./g4f/local/models/

You can get a list of the current supported models by running:

from g4f.local import LocalClient

client   = LocalClient()
client.list_models()

---

{
    "mistral-7b": {
        "path": "mistral-7b-openorca.gguf2.Q4_0.gguf",
        "ram": "8",
        "prompt": "<|im_start|>user\n%1<|im_end|>\n<|im_start|>assistant\n",
        "system": "<|im_start|>system\nYou are MistralOrca, a large language model trained by Alignment Lab AI. For multi-step problems, write out your reasoning for each step.\n<|im_end|>"
    },
    "mistral-7b-instruct": {
        "path": "mistral-7b-instruct-v0.1.Q4_0.gguf",
        "ram": "8",
        "prompt": "[INST] %1 [/INST]",
        "system": None
    },
    "gpt4all-falcon": {
        "path": "gpt4all-falcon-newbpe-q4_0.gguf",
        "ram": "8",
        "prompt": "### Instruction:\n%1\n### Response:\n",
        "system": None
    },
    "orca-2": {
        "path": "orca-2-13b.Q4_0.gguf",
        "ram": "16",
        "prompt": None,
        "system": None
    },
    "wizardlm-13b": {
        "path": "wizardlm-13b-v1.2.Q4_0.gguf",
        "ram": "16",
        "prompt": None,
        "system": None
    },
    "nous-hermes-llama2": {
        "path": "nous-hermes-llama2-13b.Q4_0.gguf",
        "ram": "16",
        "prompt": "### Instruction:\n%1\n### Response:\n",
        "system": None
    },
    "gpt4all-13b-snoozy": {
        "path": "gpt4all-13b-snoozy-q4_0.gguf",
        "ram": "16",
        "prompt": None,
        "system": None
    },
    "mpt-7b-chat": {
        "path": "mpt-7b-chat-newbpe-q4_0.gguf",
        "ram": "8",
        "prompt": "<|im_start|>user\n%1<|im_end|>\n<|im_start|>assistant\n",
        "system": "<|im_start|>system\n- You are a helpful assistant chatbot trained by MosaicML.\n- You answer questions.\n- You are excited to be able to help the user, but will refuse to do anything that could be considered harmful to the user.\n- You are more than just an information source, you are also able to write poetry, short stories, and make jokes.<|im_end|>"
    },
    "orca-mini-3b": {
        "path": "orca-mini-3b-gguf2-q4_0.gguf",
        "ram": "4",
        "prompt": "### User:\n%1\n### Response:\n",
        "system": "### System:\nYou are an AI assistant that follows instruction extremely well. Help as much as you can.\n\n"
    },
    "replit-code-3b": {
        "path": "replit-code-v1_5-3b-newbpe-q4_0.gguf",
        "ram": "4",
        "prompt": "%1",
        "system": None
    },
    "starcoder": {
        "path": "starcoder-newbpe-q4_0.gguf",
        "ram": "4",
        "prompt": "%1",
        "system": None
    },
    "rift-coder-7b": {
        "path": "rift-coder-v0-7b-q4_0.gguf",
        "ram": "8",
        "prompt": "%1",
        "system": None
    },
    "all-MiniLM-L6-v2": {
        "path": "all-MiniLM-L6-v2-f16.gguf",
        "ram": "1",
        "prompt": None,
        "system": None
    },
    "mistral-7b-german": {
        "path": "em_german_mistral_v01.Q4_0.gguf",
        "ram": "8",
        "prompt": "USER: %1 ASSISTANT: ",
        "system": "Du bist ein hilfreicher Assistent. "
    }
}

This was just first released as of 11 March 2024, and will recieve further updates soon.