AITRADER / Amsi-fin-o1.5

Amsi-fin-o1.5

A 9B-parameter financial reasoning model fine-tuned for trading strategy research, built on the Qwen3.5-9B architecture with restored vision capabilities.

Model Description

Amsi-fin-o1.5 is a domain-specialized large language model designed for trading strategy analysis, financial reasoning, and quantitative finance research. It was created by fine-tuning a Qwen3.5-9B base (with abliterated safety layers for unrestricted financial analysis) using supervised fine-tuning on a curated financial dataset.

Key Features

• Trading strategy expertise : Explains assumptions, regime dependence, signal definitions, risk management, failure modes, and invalidation conditions
• Multimodal-ready : Vision encoder weights preserved from the base Qwen3.5-9B architecture (ViT with 1152 hidden dim, 27 layers)
• Honest uncertainty : Trained to acknowledge when evidence is weak rather than fabricating data
• 9B parameters : Efficient enough for single-GPU deployment while maintaining strong reasoning

Training Details

Training Data

Fine-tuned on a curated mix from Salesforce/FinTrain :

Supplemented with quantitative finance research from arXiv categories: q-fin.PM, q-fin.RM, q-fin.ST, q-fin.TR.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "AITRADER/Amsi-fin-o1.5"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are a trading strategy research model."},
    {"role": "user", "content": "Design a mean-reversion strategy for SPY using Bollinger Bands."}
]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=2048, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

System Prompt

The model was trained with the following system prompt:

You are a trading strategy research model. Explain assumptions, regime dependence, signal definitions, risk, failure modes, and invalidation clearly. Do not invent live market data. When evidence is weak, say so.

Limitations

• Not financial advice : This model is for research and educational purposes only
• No live data : The model does not have access to real-time market data and should not be relied upon for live trading decisions
• Knowledge cutoff : Limited to training data available at time of fine-tuning
• Vision : While vision weights are present, the model was fine-tuned on text-only financial data

License

Apache 2.0

Citation

@misc{amsi-fin-o1.5,
  title={Amsi-fin-o1.5: A Financial Reasoning Model},
  author={AITRADER},
  year={2026},
  publisher={HuggingFace},
  url={https://huggingface.co/AITRADER/Amsi-fin-o1.5}
}