We've fine-tuned starcoder2-3b with an additional 0.7 billion high-quality, code-related tokens for 3 epochs. We used DeepSpeed ZeRO 3 and Flash Attention 2 to accelerate the training process. It achieves 65.9 pass@1 on HumanEval-Python. This model operates using the Alpaca instruction format (excluding the system prompt).

Usage

Here give some examples of how to use our model:

python

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
PROMPT = """### Instruction
{instruction}
### Response
"""
instruction = <Your code instruction here>
prompt = PROMPT.format(instruction=instruction)
tokenizer = AutoTokenizer.from_pretrained("TechxGenus/starcoder2-3b-instruct")
model = AutoModelForCausalLM.from_pretrained(
    "TechxGenus/starcoder2-3b-instruct",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
inputs = tokenizer.encode(prompt, return_tensors="pt")
outputs = model.generate(input_ids=inputs.to(model.device), max_new_tokens=2048)
print(tokenizer.decode(outputs[0]))

With text-generation pipeline:

python

from transformers import pipeline
import torch
PROMPT = """### Instruction
{instruction}
### Response
"""
instruction = <Your code instruction here>
prompt = PROMPT.format(instruction=instruction)
generator = pipeline(
    model="TechxGenus/starcoder2-3b-instruct",
    task="text-generation",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
result = generator(prompt, max_length=2048)
print(result[0]["generated_text"])

Note

Model may sometimes make errors, produce misleading contents, or struggle to manage tasks that are not related to coding. It has undergone very limited testing. Additional safety testing should be performed before any real-world deployments.

flyingfishinwater/starcoder2-3b-instruct-gguf

GGUF version of starcoder2-instruct

starcoder2-instruct

Usage

Note

flyingfishinwater/
starcoder2-3b-instruct-gguf