AWQ quantized version of starcoder2-3b model.
StarCoder2-3B model is a 3B parameter model trained on 17 programming languages from
The Stack v2, with opt-out requests excluded. The model uses
Grouped Query Attention,
a context window of 16,384 tokens with
a sliding window attention of 4,096 tokens, and was trained using the
Fill-in-the-Middle objective on 3+ trillion tokens.
The model was trained on GitHub code as well as additional selected data sources such as Arxiv and Wikipedia. As such it is not an instruction model and commands like "Write a function that computes the square root." do not work well.
Here are some examples to get started with the model. You can find a script for fine-tuning in StarCoder2's
GitHub repository.
1# pip install git+https://github.com/huggingface/transformers.git # TODO: merge PR to main
2from transformers import AutoModelForCausalLM, AutoTokenizer
3
4checkpoint = "bigcode/starcoder2-3b"
5device = "cuda" # for GPU usage or "cpu" for CPU usage
6
7tokenizer = AutoTokenizer.from_pretrained(checkpoint)
8# for multiple GPUs install accelerate and do `model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map="auto")`
9model = AutoModelForCausalLM.from_pretrained(checkpoint).to(device)
10
11inputs = tokenizer.encode("def print_hello_world():", return_tensors="pt").to(device)
12outputs = model.generate(inputs)
13print(tokenizer.decode(outputs[0]))
1>>> print(f"Memory footprint: {model.get_memory_footprint() / 1e6:.2f} MB")
2Memory footprint: 12624.81 MB
1# pip install accelerate
2import torch
3from transformers import AutoTokenizer, AutoModelForCausalLM
4
5checkpoint = "bigcode/starcoder2-3b"
6tokenizer = AutoTokenizer.from_pretrained(checkpoint)
7
8# for fp16 use `torch_dtype=torch.float16` instead
9model = AutoModelForCausalLM.from_pretrained(checkpoint, device_map="auto", torch_dtype=torch.bfloat16)
10
11inputs = tokenizer.encode("def print_hello_world():", return_tensors="pt").to("cuda")
12outputs = model.generate(inputs)
13print(tokenizer.decode(outputs[0]))
1>>> print(f"Memory footprint: {model.get_memory_footprint() / 1e6:.2f} MB")
2Memory footprint: 6312.41 MB
1# pip install bitsandbytes accelerate
2from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
3
4# to use 4bit use `load_in_4bit=True` instead
5quantization_config = BitsAndBytesConfig(load_in_8bit=True)
6
7checkpoint = "bigcode/starcoder2-3b"
8tokenizer = AutoTokenizer.from_pretrained(checkpoint)
9model = AutoModelForCausalLM.from_pretrained(checkpoint, quantization_config=quantization_config)
10
11inputs = tokenizer.encode("def print_hello_world():", return_tensors="pt").to("cuda")
12outputs = model.generate(inputs)
13print(tokenizer.decode(outputs[0]))
1>>> print(f"Memory footprint: {model.get_memory_footprint() / 1e6:.2f} MB")
2# load_in_8bit
3Memory footprint: 3434.07 MB
4# load_in_4bit
5>>> print(f"Memory footprint: {model.get_memory_footprint() / 1e6:.2f} MB")
6Memory footprint: 1994.90 MB
The pretraining dataset of the model was filtered for permissive licenses and code with no license only. Nevertheless, the model can generate source code verbatim from the dataset. The code's license might require attribution and/or other specific requirements that must be respected. We provide a
search index that lets you search through the pretraining data to identify where the generated code came from, and apply the proper attribution to your code.
The model has been trained on source code from 600+ programming languages. The predominant language in source is English although other languages are also present. As such the model is capable to generate code snippets provided some context but the generated code is not guaranteed to work as intended. It can be inefficient, contain bugs or exploits. See
the paper for an in-depth discussion of the model limitations.
The model is licensed under the BigCode OpenRAIL-M v1 license agreement. You can find the full agreement
here.
