sahil2801's picture
Update README.md
b2d94bc
metadata
license: bsd-3-clause
metrics:
  - code_eval
pipeline_tag: text-generation
tags:
  - code
model-index:
  - name: instruct-codegen-16B
    results:
      - task:
          type: code-generation
        dataset:
          type: openai_humaneval
          name: HumanEval
        metrics:
          - name: pass@1
            type: pass@1
            value: 0.371
            verified: false

Model Card for instruct-codegen-16B

Instruct-codegen-16B is an instruction following codegen model based on Salesforce codegen-16B-multi , finetuned on a dataset of 250k instruction-following samples in the alpaca format.

The data was not generated using any commercial LLM api.

The model achieves a result of 37.1% pass@1 on the HumanEval benchmark.

Generation

# pip install -q transformers
from transformers import AutoModelForCausalLM, AutoTokenizer
checkpoint = "sahil2801/instruct-codegen-16B"
device = "cuda"
tokenizer = AutoTokenizer.from_pretrained(checkpoint)
model = AutoModelForCausalLM.from_pretrained(checkpoint).half().to(device)
instruction = "Write a function to scrape hacker news."
prompt = f"Below is an instruction that describes a task.\n Write a response that appropriately completes the request.\n\n ### Instruction:\n{instruction}\n\n### Response:"
inputs = tokenizer(prompt, return_tensors="pt").to(device)
outputs = model.generate(**inputs,temperature=0.3,do_sample=True,max_new_tokens=256)
print(tokenizer.decode(outputs[0],skip_special_tokens=True))