metadata

license: mit
datasets:
  - Intel/orca_dpo_pairs
model-index:
  - name: SuperAligned-Jawade
    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: 71.59
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=bhavinjawade/SuperAligned-Jawade
          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: 90.58
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=bhavinjawade/SuperAligned-Jawade
          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: 60.81
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=bhavinjawade/SuperAligned-Jawade
          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: 69.17
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=bhavinjawade/SuperAligned-Jawade
          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: 83.82
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=bhavinjawade/SuperAligned-Jawade
          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: 49.2
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=bhavinjawade/SuperAligned-Jawade
          name: Open LLM Leaderboard

SOLAR-10B-OrcaDPO-Jawade

Overview

This model card is instruction finetuned version of upstage/SOLAR-10.7B-Instruct-v1.0 model. Trained on the Intel DPO Orca dataset using LoRA. Though it should be noted SOLAR-10.7B paper states that the original model for alignment was trained on Intel ORCA DPO pairs. Retraining using DPO and LoRA shows slight (<1%) improvement on OpenLLM Leaderboard benchmarks against SOLAR 10.7B-Instruct and significant over SOLAR 10.7B

How to Use This Model

To use the model bhavinjawade/SOLAR-10B-OrcaDPO-Jawade, follow these steps:

Import and Load the Model and Tokenizer Begin by importing the model and tokenizer. Load them using the from_pretrained method.

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("bhavinjawade/SOLAR-10B-OrcaDPO-Jawade")
tokenizer = AutoTokenizer.from_pretrained("bhavinjawade/SOLAR-10B-OrcaDPO-Jawade")

Format the Prompt Format the chat input as a list of messages, each with a role ('system' or 'user') and content.

message = [
    {"role": "system", "content": "You are a helpful assistant chatbot."},
    {"role": "user", "content": "Is the universe real? or is it a simulation? whats your opinion?"}
]
prompt = tokenizer.apply_chat_template(message, add_generation_prompt=True, tokenize=False)

Create a Pipeline Set up a pipeline for text generation with the loaded model and tokenizer.

pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer
)

Generate Text Use the pipeline to generate a sequence of text based on the prompt. You can adjust parameters like temperature and top_p for different styles of responses.

sequences = pipeline(
      prompt,
      do_sample=True,
    temperature=0.7,
       top_p=0.9,
       num_return_sequences=1,
       max_length=200,
   )
 print(sequences[0]['generated_text'])

This setup allows you to utilize the capabilities of the bhavinjawade/SOLAR-10B-OrcaDPO-Jawade model for generating responses to chat inputs.

License

Type: MIT License
Details: This license permits reuse, modification, and distribution for both private and commercial purposes under the terms of the MIT License.

Model Details

Model Name: SOLAR-10.7B-Instruct-v1.0
Organization: Upstage
Training Dataset: Intel/orca_dpo_pairs
Technique Used: LoRA (Low-Rank Adaptation)

Contact Information

https://bhavinjawade.github.io

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	70.86
AI2 Reasoning Challenge (25-Shot)	71.59
HellaSwag (10-Shot)	90.58
MMLU (5-Shot)	60.81
TruthfulQA (0-shot)	69.17
Winogrande (5-shot)	83.82
GSM8k (5-shot)	49.20