metadata

license: llama2

Model Card for Model ID

dragon-llama-7b-v0 part of the dRAGon ("Delivering RAG On ...") model series, RAG-instruct trained on top of a LLama-2 base model.

DRAGON models are fine-tuned with high-quality custom instruct datasets, designed for production quality use in RAG scenarios.

Benchmark Tests

Evaluated against the benchmark test: RAG-Instruct-Benchmark-Tester
Average of 2 Test Runs with 1 point for correct answer, 0.5 point for partial correct or blank / NF, 0.0 points for incorrect, and -1 points for hallucinations.

--Accuracy Score: 99.0 correct out of 100
--Not Found Classification: 95.0%
--Boolean: 82.5%
--Math/Logic: 70.0%
--Complex Questions (1-5): 4 (Low-Medium)
--Summarization Quality (1-5): 4 (Coherent, extractive)
--Hallucinations: No hallucinations observed in test runs.

For test run results (and good indicator of target use cases), please see the files ("core_rag_test" and "answer_sheet" in this repo).

Model Description

Developed by: llmware
Model type: LLama-2
Language(s) (NLP): English
License: LLama 2 Community License Agreement
Finetuned from model: Llama-2-7B-Base

Uses

The intended use of DRAGON models is two-fold:

Provide high-quality RAG-Instruct models designed for fact-based, no "hallucination" question-answering in connection with an enterprise RAG workflow.
DRAGON models are fine-tuned on top of leading base foundation models, generally in the 6-7B+ range, and purposefully rolled-out across multiple base models to provide choices and "drop-in" replacements for RAG specific use cases.
DRAGON models were trained on the same principles as the BLING models, so generally, it should be easy to "upgrade" from a BLING model in testing to a DRAGON model in production.

Direct Use

DRAGON is designed for enterprise automation use cases, especially in knowledge-intensive industries, such as financial services, legal and regulatory industries with complex information sources.

DRAGON models have been trained for common RAG scenarios, specifically: question-answering, key-value extraction, and basic summarization as the core instruction types without the need for a lot of complex instruction verbiage - provide a text passage context, ask questions, and get clear fact-based responses.

Bias, Risks, and Limitations

Any model can provide inaccurate or incomplete information, and should be used in conjunction with appropriate safeguards and fact-checking mechanisms.

How to Get Started with the Model

The fastest way to get started with dRAGon is through direct import in transformers:

from transformers import AutoTokenizer, AutoModelForCausalLM  
tokenizer = AutoTokenizer.from_pretrained("dragon-llama-7b-v0")  
model = AutoModelForCausalLM.from_pretrained("dragon-llama-7b-v0")

Please refer to the generation_test .py files in the Files repository, which includes 200 samples and script to test the model. The generation_test_llmware_script.py includes built-in llmware capabilities for fact-checking, as well as easy integration with document parsing and actual retrieval to swap out the test set for RAG workflow consisting of business documents.

The dRAGon model was fine-tuned with a simple "<human> and <bot>" wrapper, so to get the best results, wrap inference entries as:

full_prompt = "<human>: " + my_prompt + "\n" + "<bot>:"

The BLING model was fine-tuned with closed-context samples, which assume generally that the prompt consists of two sub-parts:

Text Passage Context, and
Specific question or instruction based on the text passage

To get the best results, package "my_prompt" as follows:

my_prompt = {{text_passage}} + "\n" + {{question/instruction}}

If you are using a HuggingFace generation script:

# prepare prompt packaging used in fine-tuning process
new_prompt = "<human>: " + entries["context"] + "\n" + entries["query"] + "\n" + "<bot>:"

inputs = tokenizer(new_prompt, return_tensors="pt")  
start_of_output = len(inputs.input_ids[0])

#   temperature: set at 0.3 for consistency of output
#   max_new_tokens:  set at 100 - may prematurely stop a few of the summaries

outputs = model.generate(
        inputs.input_ids.to(device),
        eos_token_id=tokenizer.eos_token_id,
        pad_token_id=tokenizer.eos_token_id,
        do_sample=True,
        temperature=0.3,
        max_new_tokens=100,
        )

output_only = tokenizer.decode(outputs[0][start_of_output:],skip_special_tokens=True)  

#   note: due to artifact of the fine-tuning, use this post-processing with HF generation 

eot = output_only.find("<|endoftext|>")
if eot > -1:
    output_only = output_only[:eot]

Model Card Contact

Darren Oberst & llmware team