Llama3-ChatQA-1.5-8B-GGUF

• Original model: Llama3-ChatQA-1.5-8B

Description

This repo contains GGUF format model files for Llama3-ChatQA-1.5-8B.

About GGUF

GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp. Here is an incomplete list of clients and libraries that are known to support GGUF:

• llama.cpp. This is the source project for GGUF, providing both a Command Line Interface (CLI) and a server option.
• text-generation-webui, Known as the most widely used web UI, this project boasts numerous features and powerful extensions, and supports GPU acceleration.
• Ollama Ollama is a lightweight and extensible framework designed for building and running language models locally. It features a simple API for creating, managing, and executing models, along with a library of pre-built models for use in various applications​
• KoboldCpp, A comprehensive web UI offering GPU acceleration across all platforms and architectures, particularly renowned for storytelling.
• GPT4All, This is a free and open source GUI that runs locally, supporting Windows, Linux, and macOS with full GPU acceleration.
• LM Studio An intuitive and powerful local GUI for Windows and macOS (Silicon), featuring GPU acceleration.
• LoLLMS Web UI. A notable web UI with a variety of unique features, including a comprehensive model library for easy model selection.
• Faraday.dev, An attractive, user-friendly character-based chat GUI for Windows and macOS (both Silicon and Intel), also offering GPU acceleration.
• llama-cpp-python, A Python library equipped with GPU acceleration, LangChain support, and an OpenAI-compatible API server.
• candle, A Rust-based ML framework focusing on performance, including GPU support, and designed for ease of use.
• ctransformers, A Python library featuring GPU acceleration, LangChain support, and an OpenAI-compatible AI server.
• localGPT An open-source initiative enabling private conversations with documents.

Explanation of quantisation methods

Click to see details

The new methods available are:

• GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
• GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
• GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
• GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
• GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw.

How to download GGUF files

Note for manual downloaders: You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single folder.

The following clients/libraries will automatically download models for you, providing a list of available models to choose from:

• LM Studio
• LoLLMS Web UI
• Faraday.dev

In text-generation-webui

Under Download Model, you can enter the model repo: LiteLLMs/Llama3-ChatQA-1.5-8B-GGUF and below it, a specific filename to download, such as: Q4_0/Q4_0-00001-of-00009.gguf.

Then click Download.

On the command line, including multiple files at once

I recommend using the huggingface-hub Python library:

pip3 install huggingface-hub

Then you can download any individual model file to the current directory, at high speed, with a command like this:

huggingface-cli download LiteLLMs/Llama3-ChatQA-1.5-8B-GGUF Q4_0/Q4_0-00001-of-00009.gguf --local-dir . --local-dir-use-symlinks False

More advanced huggingface-cli download usage (click to read)

You can also download multiple files at once with a pattern:

huggingface-cli download LiteLLMs/Llama3-ChatQA-1.5-8B-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'

For more documentation on downloading with huggingface-cli, please see: HF -> Hub Python Library -> Download files -> Download from the CLI.

To accelerate downloads on fast connections (1Gbit/s or higher), install hf_transfer:

pip3 install huggingface_hub[hf_transfer]

And set environment variable HF_HUB_ENABLE_HF_TRANSFER to 1:

HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download LiteLLMs/Llama3-ChatQA-1.5-8B-GGUF Q4_0/Q4_0-00001-of-00009.gguf --local-dir . --local-dir-use-symlinks False

Windows Command Line users: You can set the environment variable by running set HF_HUB_ENABLE_HF_TRANSFER=1 before the download command.

## Example `llama.cpp` command

Make sure you are using llama.cpp from commit d0cee0d or later.

./main -ngl 35 -m Q4_0/Q4_0-00001-of-00009.gguf --color -c 8192 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "<PROMPT>"

Change -ngl 32 to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.

Change -c 8192 to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically. Note that longer sequence lengths require much more resources, so you may need to reduce this value.

If you want to have a chat-style conversation, replace the -p <PROMPT> argument with -i -ins

For other parameters and how to use them, please refer to the llama.cpp documentation

How to run in text-generation-webui

Further instructions can be found in the text-generation-webui documentation, here: text-generation-webui/docs/04 ‐ Model Tab.md.

How to run from Python code

You can use GGUF models from Python using the llama-cpp-python or ctransformers libraries. Note that at the time of writing (Nov 27th 2023), ctransformers has not been updated for some time and is not compatible with some recent models. Therefore I recommend you use llama-cpp-python.

How to load this model in Python code, using llama-cpp-python

For full documentation, please see: llama-cpp-python docs.

First install the package

Run one of the following commands, according to your system:

# Base ctransformers with no GPU acceleration
pip install llama-cpp-python
# With NVidia CUDA acceleration
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install llama-cpp-python
# Or with OpenBLAS acceleration
CMAKE_ARGS="-DLLAMA_BLAS=ON -DLLAMA_BLAS_VENDOR=OpenBLAS" pip install llama-cpp-python
# Or with CLBLast acceleration
CMAKE_ARGS="-DLLAMA_CLBLAST=on" pip install llama-cpp-python
# Or with AMD ROCm GPU acceleration (Linux only)
CMAKE_ARGS="-DLLAMA_HIPBLAS=on" pip install llama-cpp-python
# Or with Metal GPU acceleration for macOS systems only
CMAKE_ARGS="-DLLAMA_METAL=on" pip install llama-cpp-python
# In windows, to set the variables CMAKE_ARGS in PowerShell, follow this format; eg for NVidia CUDA:
$env:CMAKE_ARGS = "-DLLAMA_OPENBLAS=on"
pip install llama-cpp-python

Simple llama-cpp-python example code

from llama_cpp import Llama
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = Llama(
  model_path="./Q4_0/Q4_0-00001-of-00009.gguf",  # Download the model file first
  n_ctx=32768,  # The max sequence length to use - note that longer sequence lengths require much more resources
  n_threads=8,            # The number of CPU threads to use, tailor to your system and the resulting performance
  n_gpu_layers=35         # The number of layers to offload to GPU, if you have GPU acceleration available
)
# Simple inference example
output = llm(
  "<PROMPT>", # Prompt
  max_tokens=512,  # Generate up to 512 tokens
  stop=["</s>"],   # Example stop token - not necessarily correct for this specific model! Please check before using.
  echo=True        # Whether to echo the prompt
)
# Chat Completion API
llm = Llama(model_path="./Q4_0/Q4_0-00001-of-00009.gguf", chat_format="llama-2")  # Set chat_format according to the model you are using
llm.create_chat_completion(
    messages = [
        {"role": "system", "content": "You are a story writing assistant."},
        {
            "role": "user",
            "content": "Write a story about llamas."
        }
    ]
)

How to use with LangChain

Here are guides on using llama-cpp-python and ctransformers with LangChain:

• LangChain + llama-cpp-python
• LangChain + ctransformers

Original model card: Llama3-ChatQA-1.5-8B

Model Details

We introduce Llama3-ChatQA-1.5, which excels at conversational question answering (QA) and retrieval-augmented generation (RAG). Llama3-ChatQA-1.5 is developed using an improved training recipe from ChatQA paper, and it is built on top of Llama-3 base model. Specifically, we incorporate more conversational QA data to enhance its tabular and arithmetic calculation capability. Llama3-ChatQA-1.5 has two variants: Llama3-ChatQA-1.5-8B and Llama3-ChatQA-1.5-70B. Both models were originally trained using Megatron-LM, we converted the checkpoints to Hugging Face format. For more information about ChatQA, check the website!

Other Resources

Llama3-ChatQA-1.5-70B   Evaluation Data   Training Data   Retriever   Website   Paper

Benchmark Results

Results in ChatRAG Bench are as follows:

| | ChatQA-1.0-7B | Command-R-Plus | Llama3-instruct-70b | GPT-4-0613 | GPT-4-Turbo | ChatQA-1.0-70B | ChatQA-1.5-8B | ChatQA-1.5-70B | | --: | :: | :: | :: | :---: | | Doc2Dial | 37.88 | 33.51 | 37.88 | 34.16 | 35.35 | 38.90 | 39.33 | 41.26 | | QuAC | 29.69 | 34.16 | 36.96 | 40.29 | 40.10 | 41.82 | 39.73 | 38.82 | | QReCC | 46.97 | 49.77 | 51.34 | 52.01 | 51.46 | 48.05 | 49.03 | 51.40 | | CoQA | 76.61 | 69.71 | 76.98 | 77.42 | 77.73 | 78.57 | 76.46 | 78.44 | | DoQA | 41.57 | 40.67 | 41.24 | 43.39 | 41.60 | 51.94 | 49.60 | 50.67 | | ConvFinQA | 51.61 | 71.21 | 76.6 | 81.28 | 84.16 | 73.69 | 78.46 | 81.88 | | SQA | 61.87 | 74.07 | 69.61 | 79.21 | 79.98 | 69.14 | 73.28 | 83.82 | | TopioCQA | 45.45 | 53.77 | 49.72 | 45.09 | 48.32 | 50.98 | 49.96 | 55.63 | | HybriDial* | 54.51 | 46.7 | 48.59 | 49.81 | 47.86 | 56.44 | 65.76 | 68.27 | | INSCIT | 30.96 | 35.76 | 36.23 | 36.34 | 33.75 | 31.90 | 30.10 | 32.31 | | Average (all) | 47.71 | 50.93 | 52.52 | 53.90 | 54.03 | 54.14 | 55.17 | 58.25 | | Average (exclude HybriDial) | 46.96 | 51.40 | 52.95 | 54.35 | 54.72 | 53.89 | 53.99 | 57.14 |

Note that ChatQA-1.5 is built based on Llama-3 base model, and ChatQA-1.0 is built based on Llama-2 base model. ChatQA-1.5 models use HybriDial training dataset. To ensure fair comparison, we also compare average scores excluding HybriDial. The data and evaluation scripts for ChatRAG Bench can be found here.

Prompt Format

We highly recommend that you use the prompt format we provide, as follows:

when context is available

System: {System}

{Context}

User: {Question}

Assistant: {Response}

User: {Question}

Assistant:

when context is not available

System: {System}

User: {Question}

Assistant: {Response}

User: {Question}

Assistant:

The content of the system's turn (i.e., {System}) for both scenarios is as follows:

This is a chat between a user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions based on the context. The assistant should also indicate when the answer cannot be found in the context.

Note that our ChatQA-1.5 models are optimized for the capability with context, e.g., over documents or retrieved context.

How to use

take the whole document as context

This can be applied to the scenario where the whole document can be fitted into the model, so that there is no need to run retrieval over the document.

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "nvidia/Llama3-ChatQA-1.5-8B"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")

messages = [
    {"role": "user", "content": "what is the percentage change of the net income from Q4 FY23 to Q4 FY24?"}
]

document = """NVIDIA (NASDAQ: NVDA) today reported revenue for the fourth quarter ended January 28, 2024, of $22.1 billion, up 22% from the previous quarter and up 265% from a year ago.\nFor the quarter, GAAP earnings per diluted share was $4.93, up 33% from the previous quarter and up 765% from a year ago. Non-GAAP earnings per diluted share was $5.16, up 28% from the previous quarter and up 486% from a year ago.\nQ4 Fiscal 2024 Summary\nGAAP\n| $ in millions, except earnings per share | Q4 FY24 | Q3 FY24 | Q4 FY23 | Q/Q | Y/Y |\n| Revenue | $22,103 | $18,120 | $6,051 | Up 22% | Up 265% |\n| Gross margin | 76.0% | 74.0% | 63.3% | Up 2.0 pts | Up 12.7 pts |\n| Operating expenses | $3,176 | $2,983 | $2,576 | Up 6% | Up 23% |\n| Operating income | $13,615 | $10,417 | $1,257 | Up 31% | Up 983% |\n| Net income | $12,285 | $9,243 | $1,414 | Up 33% | Up 769% |\n| Diluted earnings per share | $4.93 | $3.71 | $0.57 | Up 33% | Up 765% |"""

def get_formatted_input(messages, context):
    system = "System: This is a chat between a user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions based on the context. The assistant should also indicate when the answer cannot be found in the context."
    instruction = "Please give a full and complete answer for the question."

    for item in messages:
        if item['role'] == "user":
            ## only apply this instruction for the first user turn
            item['content'] = instruction + " " + item['content']
            break

    conversation = '\n\n'.join(["User: " + item["content"] if item["role"] == "user" else "Assistant: " + item["content"] for item in messages]) + "\n\nAssistant:"
    formatted_input = system + "\n\n" + context + "\n\n" + conversation

    return formatted_input

formatted_input = get_formatted_input(messages, document)
tokenized_prompt = tokenizer(tokenizer.bos_token + formatted_input, return_tensors="pt").to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(input_ids=tokenized_prompt.input_ids, attention_mask=tokenized_prompt.attention_mask, max_new_tokens=128, eos_token_id=terminators)

response = outputs[0][tokenized_prompt.input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

run retrieval to get top-n chunks as context

This can be applied to the scenario when the document is very long, so that it is necessary to run retrieval. Here, we use our Dragon-multiturn retriever which can handle conversatinoal query. In addition, we provide a few documents for users to play with.

from transformers import AutoTokenizer, AutoModelForCausalLM, AutoModel
import torch
import json

## load ChatQA-1.5 tokenizer and model
model_id = "nvidia/Llama3-ChatQA-1.5-8B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")

## load retriever tokenizer and model
retriever_tokenizer = AutoTokenizer.from_pretrained('nvidia/dragon-multiturn-query-encoder')
query_encoder = AutoModel.from_pretrained('nvidia/dragon-multiturn-query-encoder')
context_encoder = AutoModel.from_pretrained('nvidia/dragon-multiturn-context-encoder')

## prepare documents, we take landrover car manual document that we provide as an example
chunk_list = json.load(open("docs.json"))['landrover']

messages = [
    {"role": "user", "content": "how to connect the bluetooth in the car?"}
]

### running retrieval
## convert query into a format as follows:
## user: {user}\nagent: {agent}\nuser: {user}
formatted_query_for_retriever = '\n'.join([turn['role'] + ": " + turn['content'] for turn in messages]).strip()

query_input = retriever_tokenizer(formatted_query_for_retriever, return_tensors='pt')
ctx_input = retriever_tokenizer(chunk_list, padding=True, truncation=True, max_length=512, return_tensors='pt')
query_emb = query_encoder(**query_input).last_hidden_state[:, 0, :]
ctx_emb = context_encoder(**ctx_input).last_hidden_state[:, 0, :]

## Compute similarity scores using dot product and rank the similarity
similarities = query_emb.matmul(ctx_emb.transpose(0, 1)) # (1, num_ctx)
ranked_results = torch.argsort(similarities, dim=-1, descending=True) # (1, num_ctx)

## get top-n chunks (n=5)
retrieved_chunks = [chunk_list[idx] for idx in ranked_results.tolist()[0][:5]]
context = "\n\n".join(retrieved_chunks)

### running text generation
formatted_input = get_formatted_input(messages, context)
tokenized_prompt = tokenizer(tokenizer.bos_token + formatted_input, return_tensors="pt").to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = model.generate(input_ids=tokenized_prompt.input_ids, attention_mask=tokenized_prompt.attention_mask, max_new_tokens=128, eos_token_id=terminators)

response = outputs[0][tokenized_prompt.input_ids.shape[-1]:]
print(tokenizer.decode(response, skip_special_tokens=True))

Correspondence to

Zihan Liu (zihanl@nvidia.com), Wei Ping (wping@nvidia.com)

Citation

@article{liu2024chatqa,
  title={ChatQA: Surpassing GPT-4 on Conversational QA and RAG},
  author={Liu, Zihan and Ping, Wei and Roy, Rajarshi and Xu, Peng and Lee, Chankyu and Shoeybi, Mohammad and Catanzaro, Bryan},
  journal={arXiv preprint arXiv:2401.10225},
  year={2024}}

License

The use of this model is governed by the META LLAMA 3 COMMUNITY LICENSE AGREEMENT