Mxode's picture
Update README.md
62b7e50 verified
|
raw
history blame
2.27 kB
metadata
license: gpl-3.0
language:
  - en
pipeline_tag: text2text-generation

NanoLM-70M-Instruct-v1

English | 简体中文

Introduction

In order to explore the potential of small models, I have attempted to build a series of them, which are available in the NanoLM Collections.

This is NanoLM-70M-Instruct-v1. The model currently supports English only.

Model Details

Nano LMs Non-emb Params Arch Layers Dim Heads Seq Len
25M 15M MistralForCausalLM 12 312 12 2K
70M 42M LlamaForCausalLM 12 576 9 2K
0.3B 180M Qwen2ForCausalLM 12 896 14 4K
1B 840M Qwen2ForCausalLM 18 1536 12 4K

The tokenizer and model architecture of NanoLM-70M-Instruct-v1 are the same as SmolLM-135M, but the number of layers has been reduced from 30 to 12.

Essentially, it is a pure LLaMA architecture, specifically LlamaForCausalLM.

As a result, NanoLM-70M-Instruct-v1 has only 70 million parameters.

Despite this, NanoLM-70M-Instruct-v1 still demonstrates instruction-following capabilities.

How to use

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = 'Mxode/NanoLM-70M-Instruct-v1'

model = AutoModelForCausalLM.from_pretrained(model_path).to('cuda:0', torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained(model_path)


text = "Why is it important for entrepreneurs to prioritize financial management?"
prompt = tokenizer.apply_chat_template(
    [
        {'role': 'system', 'content': 'You are a helpful assistant.'},
        {'role': 'user', 'content': text}
    ],
    add_generation_prompt=True,
    tokenize=True,
    return_tensors='pt'
).to('cuda:0')


outputs = model.generate(
    prompt,
    max_new_tokens=1024,
    do_sample=True,
    temperature=0.7,
    repetition_penalty=1.1,
    eos_token_id=tokenizer.eos_token_id,
)
response = tokenizer.decode(outputs[0])
print(response)