The Yi series models are large language models trained from scratch by developers at 01.AI.

While benchmarking open-source models, we have observed a disparity between the results generated by our pipeline and those reported in public sources (e.g. OpenCompass). Upon conducting a more in-depth investigation of this difference, we have discovered that various models may employ different prompts, post-processing strategies, and sampling techniques, potentially resulting in significant variations in the outcomes. Our prompt and post-processing strategy remains consistent with the original benchmark, and greedy decoding is employed during evaluation without any post-processing for the generated content. For scores that were not reported by the original authors (including scores reported with different settings), we try to get results with our pipeline.

To evaluate the model's capability extensively, we adopted the methodology outlined in Llama2. Specifically, we included PIQA, SIQA, HellaSwag, WinoGrande, ARC, OBQA, and CSQA to assess common sense reasoning. SquAD, QuAC, and BoolQ were incorporated to evaluate reading comprehension. CSQA was exclusively tested using a 7-shot setup, while all other tests were conducted with a 0-shot configuration. Additionally, we introduced GSM8K (8-shot@1), MATH (4-shot@1), HumanEval (0-shot@1), and MBPP (3-shot@1) under the category "Math & Code". Due to technical constraints, we did not test Falcon-180 on QuAC and OBQA; the score is derived by averaging the scores on the remaining tasks. Since the scores for these two tasks are generally lower than the average, we believe that Falcon-180B's performance was not underestimated.

Feel free to create an issue if you encounter any problem when using the Yi series models.

The best approach to try the Yi series models is through Docker with GPUs. We provide the following docker images to help you get started.

• registry.lingyiwanwu.com/ci/01-ai/yi:latest

Note that the latest tag always points to the latest code in the main branch. To test a stable version, please replace it with a specific tag.

If you prefer to try out with your local development environment. First, create a virtual environment and clone this repo. Then install the dependencies with pip install -r requirements.txt. For the best performance, we recommend you also install the latest version (>=2.3.3) of flash-attention.

By default, the model weights and tokenizer will be downloaded from HuggingFace automatically in the next step. You can also download them manually from the following places:

• ModelScope
• WiseModel (Search for Yi)
• Mirror site (remember to extract the content with tar)
  • Yi-6B.tar
  • Yi-6B-200K.tar
  • Yi-34B.tar
  • Yi-34B-200K.tar

python demo/text_generation.py

To reuse the downloaded models in the previous step, you can provide the extra --model argument:

python demo/text_generation.py  --model /path/to/model

Or if you'd like to get your hands dirty:

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("01-ai/Yi-34B", device_map="auto", torch_dtype="auto", trust_remote_code=True)
tokenizer = AutoTokenizer.from_pretrained("01-ai/Yi-34B", trust_remote_code=True)
inputs = tokenizer("There's a place where time stands still. A place of breath taking wonder, but also", return_tensors="pt")
max_length = 256

outputs = model.generate(
    inputs.input_ids.cuda(),
    max_length=max_length,
    eos_token_id=tokenizer.eos_token_id,
    do_sample=True,
    repetition_penalty=1.3,
    no_repeat_ngram_size=5,
    temperature=0.7,
    top_k=40,
    top_p=0.8,
)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

For more advanced usage, please refer to the doc.

bash finetune/scripts/run_sft_Yi_6b.sh

Once finished, you can compare the finetuned model and the base model with the following command:

bash finetune/scripts/run_eval.sh

For more advanced usage like fine-tuning based on your custom data, please refer the doc.

python quantization/gptq/quant_autogptq.py \
  --model /base_model                      \
  --output_dir /quantized_model            \
  --trust_remote_code

Once finished, you can then evaluate the resulting model as follows:

python quantization/gptq/eval_quantized_model.py \
  --model /quantized_model                       \
  --trust_remote_code

For a more detailed explanation, please read the doc

python quantization/awq/quant_autoawq.py \
  --model /base_model                      \
  --output_dir /quantized_model            \
  --trust_remote_code

Once finished, you can then evaluate the resulted model as follows:

python quantization/awq/eval_quantized_model.py \
  --model /quantized_model                       \
  --trust_remote_code

For more detailed explanation, please read the doc

We use data compliance checking algorithms during the training process, to ensure the compliance of the trained model to the best of our ability. Due to complex data and the diversity of language model usage scenarios, we cannot guarantee that the model will generate correct, and reasonable output in all scenarios. Please be aware that there is still a risk of the model producing problematic outputs. We will not be responsible for any risks and issues resulting from misuse, misguidance, illegal usage, and related misinformation, as well as any associated data security concerns.

The source code in this repo is licensed under the Apache 2.0 license. The Yi series models are fully open for academic research and free commercial usage with permission via applications. All usage must adhere to the Model License Agreement 2.0. To apply for the official commercial license, please contact us (yi@01.ai).