XelotX/DeepSeek-V2-Chat-0628-XelotX-Quants

GGUF version made with llama.cpp 705b7ec

Original model deepseek-ai/DeepSeek-V2-Chat-0628

---

Model Download | Evaluation Results | Model Architecture | API Platform | License | Citation

Paper Link👁️

DeepSeek-V2-Chat-0628

1. Introduction

DeepSeek-V2-Chat-0628 is an improved version of DeepSeek-V2-Chat. For model details, please visit DeepSeek-V2 page for more information.

DeepSeek-V2-Chat-0628 has achieved remarkable performance on the LMSYS Chatbot Arena Leaderboard:

Overall Ranking: #11, outperforming all other open-source models.

Coding Arena Ranking: #3, showcasing exceptional capabilities in coding tasks.

Hard Prompts Arena Ranking: #3, demonstrating strong performance on challenging prompts.

2. Improvement

Compared to the previous version DeepSeek-V2-Chat, the new version has made the following improvements:

Benchmark	DeepSeek-V2-Chat	DeepSeek-V2-Chat-0628	Improvement
HumanEval	81.1	84.8	+3.7
MATH	53.9	71.0	+17.1
BBH	79.7	83.4	+3.7
IFEval	63.8	77.6	+13.8
Arena-Hard	41.6	68.3	+26.7
JSON Output (Internal)	78	85	+7

Furthermore, the instruction following capability in the "system" area has been optimized, significantly enhancing the user experience for immersive translation, RAG, and other tasks.

3. How to run locally

To utilize DeepSeek-V2-Chat-0628 in BF16 format for inference, 80GB*8 GPUs are required.

Inference with Huggingface's Transformers

You can directly employ Huggingface's Transformers for model inference.

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig

model_name = "deepseek-ai/DeepSeek-V2-Chat-0628"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
# `max_memory` should be set based on your devices
max_memory = {i: "75GB" for i in range(8)}
# `device_map` cannot be set to `auto`
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True, device_map="sequential", torch_dtype=torch.bfloat16, max_memory=max_memory, attn_implementation="eager")
model.generation_config = GenerationConfig.from_pretrained(model_name)
model.generation_config.pad_token_id = model.generation_config.eos_token_id

messages = [
    {"role": "user", "content": "Write a piece of quicksort code in C++"}
]
input_tensor = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
outputs = model.generate(input_tensor.to(model.device), max_new_tokens=100)

result = tokenizer.decode(outputs[0][input_tensor.shape[1]:], skip_special_tokens=True)
print(result)

The complete chat template can be found within tokenizer_config.json located in the huggingface model repository.

Note: The chat template has been updated compared to the previous DeepSeek-V2-Chat version.

An example of chat template is as belows:

<｜begin▁of▁sentence｜><｜User｜>{user_message_1}<｜Assistant｜>{assistant_message_1}<｜end▁of▁sentence｜><｜User｜>{user_message_2}<｜Assistant｜>

You can also add an optional system message:

<｜begin▁of▁sentence｜>{system_message}

<｜User｜>{user_message_1}<｜Assistant｜>{assistant_message_1}<｜end▁of▁sentence｜><｜User｜>{user_message_2}<｜Assistant｜>

Inference with vLLM (recommended)

To utilize vLLM for model inference, please merge this Pull Request into your vLLM codebase: https://github.com/vllm-project/vllm/pull/4650.

from transformers import AutoTokenizer
from vllm import LLM, SamplingParams

max_model_len, tp_size = 8192, 8
model_name = "deepseek-ai/DeepSeek-V2-Chat-0628"
tokenizer = AutoTokenizer.from_pretrained(model_name)
llm = LLM(model=model_name, tensor_parallel_size=tp_size, max_model_len=max_model_len, trust_remote_code=True, enforce_eager=True)
sampling_params = SamplingParams(temperature=0.3, max_tokens=256, stop_token_ids=[tokenizer.eos_token_id])

messages_list = [
    [{"role": "user", "content": "Who are you?"}],
    [{"role": "user", "content": "Translate the following content into Chinese directly: DeepSeek-V2 adopts innovative architectures to guarantee economical training and efficient inference."}],
    [{"role": "user", "content": "Write a piece of quicksort code in C++."}],
]

prompt_token_ids = [tokenizer.apply_chat_template(messages, add_generation_prompt=True) for messages in messages_list]

outputs = llm.generate(prompt_token_ids=prompt_token_ids, sampling_params=sampling_params)

generated_text = [output.outputs[0].text for output in outputs]
print(generated_text)

4. License

This code repository is licensed under the MIT License. The use of DeepSeek-V2 Base/Chat models is subject to the Model License. DeepSeek-V2 series (including Base and Chat) supports commercial use.

5. Citation

@misc{deepseekv2,
      title={DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model}, 
      author={DeepSeek-AI},
      year={2024},
      eprint={2405.04434},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

6. Contact

If you have any questions, please raise an issue or contact us at [email protected].