README.md

metadata

tags:
  - gguf
  - qwen2
  - finance
  - news-analysis
  - stock-market
base_model: Qwen/Qwen2.5-0.5B-Instruct
license: apache-2.0

qwen-0.5b-stocks-news-v2 (GGUF)

GGUF quantized versions of qwen-0.5b-stocks-news-v2.

Model Description

Fine-tuned Qwen2.5-0.5B-Instruct for financial news analysis. Outputs structured JSON with:

• impactTerm: LONGTERM / SHORTTERM
• impactScope: GLOBAL / INDUSTRY / STOCK
• scopeItems: list of affected stocks/industries
• highlights: key information summary
• sentiment: POSITIVE / NEGATIVE / NEUTRAL

Available Quantizations

Quantization	Size	Quality	Use Case
F16	~1 GB	Highest	Reference / GPU inference
Q8_0	~530 MB	Very High	Desktop / server inference
Q4_K_M	~320 MB	Good	Mobile / edge / low-RAM

Usage

With llama.cpp

./llama-cli -m qwen-0.5b-stocks-news-v2-q4_k_m.gguf \
  -p '<|im_start|>system
You are a financial news AI. Analyze the given news article and output its scope, impact, highlights, and sentiment in JSON format. /no_think<|im_end|>
<|im_start|>user
RBI holds repo rate at 6.5% in MPC meeting<|im_end|>
<|im_start|>assistant
' \
  -n 256 --temp 0.7

With Ollama

Create a Modelfile:

FROM ./qwen-0.5b-stocks-news-v2-q4_k_m.gguf

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

SYSTEM "You are a financial news AI. Analyze the given news article and output its scope, impact, highlights, and sentiment in JSON format. /no_think"

PARAMETER temperature 0.7
PARAMETER top_p 0.8
PARAMETER stop <|im_end|>

Then:

ollama create qwen-0.5b-stocks-news-v2 -f Modelfile
ollama run qwen-0.5b-stocks-news-v2 "Infosys Q3 profit rises 11%"