metadata

license: apache-2.0
datasets:
  - ariaattarml/verified-reasoning-cot-gpqa-mmlu-pro
base_model:
  - Qwen/Qwen2.5-72B-Instruct

S0 Series Models

Overview

The S0 v0.1 model by TensorStax is an early preview line of open reasoning models, specifically trained using process supervision on synthetic data. These models are designed to excel at general reasoning tasks while maintaining transparency in their thought processes.

Future Specialized Releases

We plan to expand the S0 series with specialized variants optimized for specific domains:

Code Generation
Query Generation
Long horizon agency

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "ariaattarml/TensorStax-72B-S0-0.1"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,
    device_map="auto",
)

alpaca_prompt = """
### Instruction:
{}

### Input:
{}

### Response:
{}"""

system_prompt = """
You are an intelligent reasoning AI assistant
"""
user_prompt = """
As of 2017, how many of the world's 1-year-old children today have been vaccinated against some disease?

Options: ['30%', '60%', '10%', '90%', '80%', '40%', '100%', '50%', 'N/A', 'N/A']
"""

inputs = tokenizer(
[
    alpaca_prompt.format(
        system_prompt,
        user_prompt,
        "",
    )
], return_tensors = "pt").to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)

_ = model.generate(
    **inputs,
    streamer = text_streamer,
    max_new_tokens = 8000,
    temperature = 1.0
)