This is the Qwen/Qwen2-1.5B-Instruct model with the Replete-AI/Adapter_For_Replete-Coder-Qwen2-1.5b applied on top of it.
This is mostly an experinment to see how the model would perform.
Links to the oringal model and adapter are bellow:
Orginal model:
Adapter:
Finetuned by: Rombodawg
Although Replete-Coder has amazing coding capabilities, its trained on vaste amount of non-coding data, fully cleaned and uncensored. Dont just use it for coding, use it for all your needs! We are truly trying to make the GPT killer!
Thank you to TensorDock for sponsoring Replete-Coder-llama3-8b and Replete-Coder-Qwen2-1.5b you can check out their website for cloud compute rental bellow.
Replete-Coder-Qwen2-1.5b is a general purpose model that is specially trained in coding in over 100 coding languages. The data used to train the model contains 25% non-code instruction data and 75% coding instruction data totaling up to 3.9 million lines, roughly 1 billion tokens, or 7.27gb of instruct data. The data used to train this model was 100% uncensored, then fully deduplicated, before training happened.
The Replete-Coder models (including Replete-Coder-llama3-8b and Replete-Coder-Qwen2-1.5b) feature the following:
Notice: Replete-Coder series of models are fine-tuned on a context window of 8192 tokens. Performance past this context window is not guaranteed.
You can find the 25% non-coding instruction below:
And the 75% coding specific instruction data below:
These two datasets were combined to create the final dataset for training, which is linked below:
<|im_start|>system
{}<|im_end|>
<|im_start|>user
{}<|im_end|>
<|im_start|>assistant
{}
Note: The system prompt varies in training data, but the most commonly used one is:
Below is an instruction that describes a task, Write a response that appropriately completes the request.
End token:
<|endoftext|>
Thank you to the community for your contributions to the Replete-AI/code_bagel_hermes-2.5 dataset. Without the participation of so many members making their datasets free and open source for any to use, this amazing AI model wouldn't be possible.
Extra special thanks to Teknium for the Open-Hermes-2.5 dataset and jondurbin for the bagel dataset and the naming idea for the code_bagel series of datasets. You can find both of their huggingface accounts linked below:
Another special thanks to unsloth for being the main method of training for Replete-Coder. Bellow you can find their github, as well as the special Replete-Ai secret sause (Unsloth + Qlora + Galore) colab code document that was used to train this model.
Qwen2 is the new series of Qwen large language models. For Qwen2, we release a number of base language models and instruction-tuned language models ranging from 0.5 to 72 billion parameters, including a Mixture-of-Experts model. This repo contains the instruction-tuned 1.5B Qwen2 model.
Compared with the state-of-the-art opensource language models, including the previous released Qwen1.5, Qwen2 has generally surpassed most opensource models and demonstrated competitiveness against proprietary models across a series of benchmarks targeting for language understanding, language generation, multilingual capability, coding, mathematics, reasoning, etc.
For more details, please refer to our blog, GitHub, and Documentation.
Qwen2 is a language model series including decoder language models of different model sizes. For each size, we release the base language model and the aligned chat model. It is based on the Transformer architecture with SwiGLU activation, attention QKV bias, group query attention, etc. Additionally, we have an improved tokenizer adaptive to multiple natural languages and codes.
We pretrained the models with a large amount of data, and we post-trained the models with both supervised finetuning and direct preference optimization.
The code of Qwen2 has been in the latest Hugging face transformers and we advise you to install transformers>=4.37.0, or you might encounter the following error:
KeyError: 'qwen2'
Here provides a code snippet with apply_chat_template to show you how to load the tokenizer and model and how to generate contents.
from transformers import AutoModelForCausalLM, AutoTokenizer
device = "cuda" # the device to load the model onto
model = AutoModelForCausalLM.from_pretrained(
"Qwen/Qwen2-1.5B-Instruct",
torch_dtype="auto",
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2-1.5B-Instruct")
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(device)
generated_ids = model.generate(
model_inputs.input_ids,
max_new_tokens=512
)
generated_ids = [
output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]
response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
We briefly compare Qwen2-1.5B-Instruct with Qwen1.5-1.8B-Chat. The results are as follows:
| Datasets | Qwen1.5-0.5B-Chat | Qwen2-0.5B-Instruct | Qwen1.5-1.8B-Chat | **
Qwen2-1.5B-Instruct** | | :--- | :---: | :---: | :---: | :---: | | MMLU | 35.0 | 37.9 | 43.7 | 52.4 | | HumanEval | 9.1 | 17.1 | 25.0 | 37.8 | | GSM8K | 11.3 | 40.1 | 35.3 | 61.6 | | C-Eval | 37.2 | 45.2 | 55.3 | 63.8 | | IFEval (Prompt Strict-Acc.) | 14.6 | 20.0 | 16.8 | 29.0 |
If you find our work helpful, feel free to give us a cite.
@article{qwen2,
title={Qwen2 Technical Report},
year={2024}
}