AIGym's picture
Adding Evaluation Results (#1)
0cb76ea verified
metadata
license: apache-2.0
tags:
  - finetuned
pipeline_tag: text-generation
model-index:
  - name: deepseek-coder-6.7b-chat
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 36.01
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AIGym/deepseek-coder-6.7b-chat
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 53.74
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AIGym/deepseek-coder-6.7b-chat
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 38.22
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AIGym/deepseek-coder-6.7b-chat
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 42.94
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AIGym/deepseek-coder-6.7b-chat
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 57.54
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AIGym/deepseek-coder-6.7b-chat
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 16.98
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=AIGym/deepseek-coder-6.7b-chat
          name: Open LLM Leaderboard

deepseek-coder-6.7B-chat

It was created by starting with the deepseek-coder-6.7B and training it on the open assistant dataset. We have attached the wandb report in pdf form to view the training run at a glance.

Reson

This model was fine tned to allow it to follow direction and is a steeping stone to further training, but still would be good for asking qestions about code.

How to use

You will need the transformers>=4.31

from transformers import AutoTokenizer
import transformers 
import torch
model = "AIGym/deepseek-coder-6.7b-chat"
tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

prompt = "What are the values in open source projects?"
formatted_prompt = (
    f"### Human: {prompt}### Assistant:"
)


sequences = pipeline(
    formatted_prompt,
    do_sample=True,
    top_k=50,
    top_p = 0.7,
    num_return_sequences=1,
    repetition_penalty=1.1,
    max_new_tokens=500,
)
for seq in sequences:
    print(f"Result: {seq['generated_text']}")

Referrals

Run Pod - This is who I use to train th emodels on huggingface. If you use it we both get free crdits. - Visit Runpod's Website!

Paypal - If you want to leave a tip, it is appecaheted. - Visit My Paypal!

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 40.90
AI2 Reasoning Challenge (25-Shot) 36.01
HellaSwag (10-Shot) 53.74
MMLU (5-Shot) 38.22
TruthfulQA (0-shot) 42.94
Winogrande (5-shot) 57.54
GSM8k (5-shot) 16.98