p988744/eland-sentiment-zh-vllm

Eland Sentiment vLLM - Chinese Multi-Domain Sentiment Analysis

A merged Qwen3-4B model fine-tuned for Chinese multi-domain sentiment analysis, optimized for vLLM deployment.

This is the full merged model (LoRA weights merged into base model) for high-throughput inference with vLLM.

Performance

Financial Domain (92.03%)

Metric	Score
Overall Sentiment	95.00%
Entity Sentiment	93.10%
Opinion Sentiment	85.00%
Agrees with Text	95.00%

Multi-Domain (86.85%)

Metric	Score
Overall Sentiment	71.00%
Entity Sentiment	78.57%
Opinion Sentiment	97.83%
Agrees with Text	100.00%

⚠️ Important: System Prompt Required

vLLM 部署必須提供 System Prompt，否則模型可能無法正確輸出。

與 Ollama 不同，vLLM 不會讀取 Modelfile 中的預設設定，因此必須在每次請求中包含 System Prompt。

標準 System Prompt：

你是一個專業的金融文本情感分析助手。請分析以下文本的情感，回答「正面」、「負面」或「中立」。

Usage with vLLM

Basic Usage

from vllm import LLM, SamplingParams

# Load model
llm = LLM(model="p988744/eland-sentiment-zh-vllm")

# Define sampling parameters
sampling_params = SamplingParams(
    temperature=0.1,
    top_p=0.9,
    max_tokens=10
)

# Create prompt
prompt = """<|im_start|>system
你是一個專業的金融文本情感分析助手。請分析以下文本的整體情感，回答「正面」、「負面」或「中立」。<|im_end|>
<|im_start|>user
台積電今日股價大漲，市場看好AI需求持續成長。<|im_end|>
<|im_start|>assistant
"""

# Generate
outputs = llm.generate([prompt], sampling_params)
print(outputs[0].outputs[0].text)  # Expected: 正面

Batch Processing

from vllm import LLM, SamplingParams

llm = LLM(model="p988744/eland-sentiment-zh-vllm")
sampling_params = SamplingParams(temperature=0.1, max_tokens=10)

# Multiple texts
texts = [
    "台積電營收創新高",
    "投資人對後市持觀望態度",
    "公司宣布大幅裁員"
]

prompts = []
for text in texts:
    prompt = f"""<|im_start|>system
你是一個專業的金融文本情感分析助手。請分析以下文本的整體情感，回答「正面」、「負面」或「中立」。<|im_end|>
<|im_start|>user
{text}<|im_end|>
<|im_start|>assistant
"""
    prompts.append(prompt)

outputs = llm.generate(prompts, sampling_params)
for text, output in zip(texts, outputs):
    print(f"{text} -> {output.outputs[0].text}")

OpenAI-Compatible Server

# Start vLLM server
vllm serve p988744/eland-sentiment-zh-vllm --port 8000

# Query with curl
curl http://localhost:8000/v1/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "p988744/eland-sentiment-zh-vllm",
    "prompt": "<|im_start|>system\n你是一個專業的金融文本情感分析助手。請分析以下文本的整體情感，回答「正面」、「負面」或「中立」。<|im_end|>\n<|im_start|>user\n台積電股價大漲<|im_end|>\n<|im_start|>assistant\n",
    "max_tokens": 10,
    "temperature": 0.1
  }'

Task Prompts

Overall Sentiment:

System: 你是一個專業的金融文本情感分析助手。請分析以下文本的整體情感，回答「正面」、「負面」或「中立」。
User: [your text]

Entity Sentiment:

System: 你是一個專業的金融文本情感分析助手。請分析以下文本中對「{entity}」的情感，回答「正面」、「負面」或「中立」。
User: [your text]

Opinion Sentiment:

System: 你是一個專業的金融文本情感分析助手。請判斷以下觀點的情感傾向，回答「正面」、「負面」或「中立」。
User: 文本：[text]
觀點：[opinion]

Model Variants

Version	Repository	Use Case
LoRA Adapter	p988744/eland-sentiment-zh	HuggingFace + PEFT
GGUF	p988744/eland-sentiment-zh-gguf	Ollama / llama.cpp
Full Merged	p988744/eland-sentiment-zh-vllm	vLLM (this repo)

Model Details

Parameter	Value
Base Model	Qwen/Qwen3-4B
Parameters	4.05B
dtype	bfloat16
Model Size	~8GB
Context Length	2048

Training Details

Parameter	Value
Method	LoRA (merged)
LoRA Rank	32
LoRA Alpha	64
Epochs	8
Learning Rate	1e-5

Dataset

Combined multi-domain dataset:

• Financial: 1,887 training samples (Taiwan stock market)
• Multi-domain: 600 training samples (product, brand, organization, social)
• Total: 2,487 training samples

Requirements

pip install vllm>=0.4.0

License

Apache 2.0