README.md

---
library_name: transformers
tags:
- time series
- multimodal
- TimeSeries-Text-to-Text
license: apache-2.0
---

# Mists-7B-v01-not-trained

Mists(**Mis**tral **T**ime **S**eries) is a multimodal model that combines language and time series model.  
This model is based on the following models:  
 - [mistralai/Mistral-7B-Instruct-v0.3](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3)
 - [HachiML/MOMENT-1-large-embedding-v0.1](https://huggingface.co/HachiML/MOMENT-1-large-embedding-v0.1) (an embedding model derived from [AutonLab/MOMENT-1-large](https://huggingface.co/AutonLab/MOMENT-1-large))

This is an experimental model.
Since the adapter has not been trained, the model is not yet suitable for use.

## How to use

```Python
!pip install accelerate
```

```Python
from transformers import AutoProcessor, AutoModel
import torch

model_id = "HachiML/Mists-7B-v01-not-trained"

processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
model = AutoModel.from_pretrained(
    model_id,
    torch_dtype=torch.float32,
    low_cpu_mem_usage=True,
    device_map="auto",
    trust_remote_code=True,
)
```

```Python
import pandas as pd
import torch

hist_ndaq = pd.DataFrame("nasdaq_price_history.csv")
time_series_data = hist_ndaq[["Open", "High", "Low", "Close", "Volume"]].iloc[:512]

prompt = "USER: <time_series>\nWhat are the features of this data?\nASSISTANT:"
inputs = processor(prompt, time_series_data, return_tensors='pt')

device = "cuda" if torch.cuda.is_available() else "cpu"
for key, item in inputs.items():
    inputs[key] = inputs[key].to(device)

output = model.generate(**inputs, max_new_tokens=200, do_sample=False)
print(processor.decode(output[0], skip_special_tokens=False))
```