p988744/eland-sentiment-zh-gguf

Eland Sentiment GGUF - Chinese Financial Sentiment Analysis

GGUF quantized versions of the Eland Sentiment model for Ollama and llama.cpp deployment.

Available Files

File	Size	Quantization	Use Case
eland-sentiment-balanced-q8_0.gguf	4.0 GB	Q8_0	Recommended - Best balance of size and quality
eland-sentiment-balanced-f16.gguf	7.5 GB	F16	Full precision, larger but highest quality

Performance (Balanced Training v3)

Metric	Score
Macro Average	89.80%
正面 (Positive)	91.19%
負面 (Negative)	88.69%
中立 (Neutral)	89.52%

Usage with Ollama

Quick Start

# Download GGUF and Modelfile
wget https://huggingface.co/p988744/eland-sentiment-zh-gguf/resolve/main/eland-sentiment-balanced-q8_0.gguf
wget https://huggingface.co/p988744/eland-sentiment-zh-gguf/resolve/main/Modelfile

# Create Ollama model
ollama create eland-sentiment-zh -f Modelfile

# Run inference
ollama run eland-sentiment-zh "台積電今日股價大漲，市場看好AI需求持續成長。"

Custom Modelfile

The included Modelfile is configured for sentiment analysis:

FROM ./eland-sentiment-balanced-q8_0.gguf

TEMPLATE """{{- if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{ end }}<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""

PARAMETER stop "<|im_start|>"
PARAMETER stop "<|im_end|>"
PARAMETER temperature 0.1
PARAMETER top_p 0.9

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

LICENSE """Apache 2.0"""

API Usage

# Start Ollama server (if not running)
ollama serve

# Query via API
curl http://localhost:11434/api/generate -d '{
  "model": "eland-sentiment-zh",
  "prompt": "台積電股價大漲",
  "stream": false
}'

Python with Ollama

import ollama

response = ollama.generate(
    model='eland-sentiment-zh',
    prompt='台積電今日股價大漲，市場看好AI需求持續成長。'
)
print(response['response'])  # Expected: 正面

Usage with llama.cpp

CLI

# Download GGUF
wget https://huggingface.co/p988744/eland-sentiment-zh-gguf/resolve/main/eland-sentiment-balanced-q8_0.gguf

# Run inference
./llama-cli -m eland-sentiment-balanced-q8_0.gguf \
  -p "<|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<|im_end|>
<|im_start|>assistant
" \
  -n 10 \
  --temp 0.1

llama-cpp-python

from llama_cpp import Llama

llm = Llama(
    model_path="eland-sentiment-balanced-q8_0.gguf",
    n_ctx=2048,
    n_threads=4
)

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
"""

output = llm(prompt, max_tokens=10, temperature=0.1)
print(output['choices'][0]['text'])  # Expected: 正面

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 (this repo)
Full Merged	p988744/eland-sentiment-zh-vllm	vLLM

Model Details

Parameter	Value
Base Model	Qwen/Qwen3-4B
Parameters	4.05B
Tensors	398
Context Length	2048

Dataset

Trained on p988744/eland-sentiment-zh-data:

• 1,887 training samples (balanced: 33.3% each class)
• 300 test samples
• Taiwan stock market forum and news text

License

Apache 2.0