Instructions to use egg1234/eggAI with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use egg1234/eggAI with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="egg1234/eggAI")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("egg1234/eggAI")
model = AutoModelForCausalLM.from_pretrained("egg1234/eggAI", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use egg1234/eggAI with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf egg1234/eggAI:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf egg1234/eggAI:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf egg1234/eggAI:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf egg1234/eggAI:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf egg1234/eggAI:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf egg1234/eggAI:Q4_K_M

Use Docker

docker model run hf.co/egg1234/eggAI:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use egg1234/eggAI with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "egg1234/eggAI"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "egg1234/eggAI",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/egg1234/eggAI:Q4_K_M

SGLang

How to use egg1234/eggAI with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "egg1234/eggAI" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "egg1234/eggAI",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "egg1234/eggAI" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "egg1234/eggAI",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use egg1234/eggAI with Ollama:
```
ollama run hf.co/egg1234/eggAI:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use egg1234/eggAI with Docker Model Runner:
```
docker model run hf.co/egg1234/eggAI:Q4_K_M
```

Lemonade

How to use egg1234/eggAI with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull egg1234/eggAI:Q4_K_M

Run and chat with the model

lemonade run user.eggAI-Q4_K_M

List all available models

lemonade list

Atomic Chat

首先：
原模型：Qwen3.5-0.8B
许可证：Apache-2.0
改动：二次微调权重，加强python编程能力

------------

一、这份东西是什么**

eggAI 是一个约0.75B（由Qwen3.5-0.8B剥离多模态视觉理解得来）参数的中文本地助手。
它只做三件事：写 Python 代码、解释报错、日常问答（闲聊、身份、简单推理）。

二、目录结构

model.safetensors（约 1.4 GB）完整权重，配合 config.json / tokenizer* 直接加载
chat_template.jinja 对话模板（极简版，含默认身份）
GGUF文件夹中还有可直接给客户端用的量化版本：
eggAI-Q4_K_M.gguf（约 505 MB）

三、使用方法

方式 1：客户端（最简单）

使用llama.cpp或类似软件直接加载GGUF文件夹中的的 eggAI-Q4_K_M.gguf。

建议 temperature=0.7~0.9

方式 2：llama.cpp 命令行

llama-cli -m eggAI-Q4_K_M.gguf -cnv --jinja --temp 0.3 -ngl 99

方式 3：Python（transformers）

from transformers import AutoTokenizer, AutoModelForCausalLM
tok = AutoTokenizer.from_pretrained(r".")
m = AutoModelForCausalLM.from_pretrained(r".", dtype="float32")
msgs = [{"role": "user", "content": "你是谁？"}]
pp = tok.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True)
ids = tok(pp, return_tensors="pt", add_special_tokens=False)
out = m.generate(**ids, max_new_tokens=300, do_sample=False)
print(tok.decode(out[0][ids.input_ids.shape[1]:], skip_special_tokens=False))

四、训练数据构成

类别	数量	来源与验证方式
编程	316	Alpaca 中文代码题（全部过 py_compile）+ 题库（真实执行比对输出）
指令跟随	200	Alpaca 中文（改写/摘要/分类/提取/解释等）
数学	104	手写应用题（带 3-4 步推理）+ 题库（答案比对）
Windows 命令	28	手写标准 PowerShell / CMD 命令
闲聊（弱智吧）	42	弱智吧题目 + 手写 eggAI 口吻回答
身份	30 ×4 加权	覆盖「你是 Qwen 吗/通义吗/ChatGPT 吗」等所有变体

训练采用两阶段：先只训身份（30 条 ×5 轮，lr 3e-4）把「我是 eggAI / eggking 研发」写进权重，
再以该模型为底模训全量能力（1230 条 ×3 轮，lr 1.2e-4），避免能力数据稀释身份。

五、已知限制

数学能力弱：0.75B 参数量决定，复杂应用题（如鸡兔同笼）会算错
长上下文：训练时 max_len 1024，超长输入会截断
知识面窄：没有注入世界知识，问事实性问题容易编

六、模型训练参数

底模：Qwen3.5 架构，24 层，hidden 1024
LoRA：r=16, alpha=32, dropout=0.05，目标模块 q_proj/k_proj/v_proj/o_proj/gate_proj/up_proj/down_proj
学习率：身份阶段 3e-4，能力阶段 1.2e-4～1.5e-4
停止符：<|im_end|>(248046) + <|endoftext|>(248044)

七、评测记录

对比原qwen3.5-0.8B模型各方面小幅度领先，对比DeepSeek-R1-Distill-Qwen-1.5B-Q5_K_M部分场景领先，代码能力在no think的情况下会更好，数学受益于长CoT思维链链路能力对比同规模模型普遍更佳。

Downloads last month: 54

Safetensors

Model size

0.8B params

Tensor type

BF16

Model tree for egg1234/eggAI

Base model

Qwen/Qwen3.5-0.8B-Base

Finetuned

Qwen/Qwen3.5-0.8B

Quantized

(418)

this model