README.md

metadata

datasets:
  - codeparrot/self-instruct-starcoder
pipeline_tag: text2text-generation
metrics:
  - code_eval
library_name: transformers
tags:
  - code
model-index:
  - name: StarCoder-SelfInstruct
    results:
      - task:
          type: text-generation
        dataset:
          type: openai_humaneval
          name: InstructHumanEval
        metrics:
          - name: pass@1
            type: pass@1
            value: 0.391
            verified: false
      - task:
          type: text-generation
        dataset:
          type: openai_humaneval
          name: HumanEval
        metrics:
          - name: pass@1
            type: pass@1
            value: 0.346
            verified: false

Model Card for Self-instruct-starcoder

This model is an instruction-tuned version of ⭐️ StarCoder. The instruction dataset involved is Self-instruct-starcoder which was built by boostrapping on StarCoder's generations.

Uses

The model was fine-tuned with the following template

Question: <instruction>

Answer: <output>

If you have your model and tokenizer loaded, you can use the following code to make the model generate the right output to a given instruction

instruction = "Write a function to compute the GCD between two integers a and b"
prompt = f"Question:{instruction}\n\nAnswer:"
input_ids = tokenizer(prompt, return_tensors="pt")["input_ids"]
completion = model.generate(input_ids, max_length=200)
print(tokenizer.batch_decode(completion[:,input_ids.shape[1]:])[0])

More information

For additional information, check

• self-intruct-starcoder
• starcoder