README.md

metadata

license: apache-2.0
model-index:
  - name: Soniox-7B-v1.0
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 63.91
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=soniox/Soniox-7B-v1.0
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 82.55
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=soniox/Soniox-7B-v1.0
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 64.38
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=soniox/Soniox-7B-v1.0
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 53.84
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=soniox/Soniox-7B-v1.0
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 78.06
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=soniox/Soniox-7B-v1.0
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 56.25
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=soniox/Soniox-7B-v1.0
          name: Open LLM Leaderboard

Model Card for Soniox-7B-v1.0

Soniox 7B is a powerful large language model. Supports English and code with 8K context. Matches GPT-4 performance on some benchmarks. Built on top of Mistral 7B, enhanced with additional pre-training and fine-tuning for strong problem-solving capabilities. Apache 2.0 License. For more details, please read our blog post.

Usage in Transformers

The model is available in transformers and can be used as follows:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "soniox/Soniox-7B-v1.0"
model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=torch.float16)
tokenizer = AutoTokenizer.from_pretrained(model_path)

device = "cuda"
model.to(device)

messages = [
    {"role": "user", "content": "12 plus 21?"},
    {"role": "assistant", "content": "33."},
    {"role": "user", "content": "Five minus one?"},
]
tok_prompt = tokenizer.apply_chat_template(messages, return_tensors="pt")

model_inputs = tok_prompt.to(device)
generated_ids = model.generate(model_inputs, max_new_tokens=1000, do_sample=True)
decoded = tokenizer.batch_decode(generated_ids)
print(decoded[0])

Inference deployment

Refer to our documentation for inference with vLLM and other deployment options.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	66.50
AI2 Reasoning Challenge (25-Shot)	63.91
HellaSwag (10-Shot)	82.55
MMLU (5-Shot)	64.38
TruthfulQA (0-shot)	53.84
Winogrande (5-shot)	78.06
GSM8k (5-shot)	56.25