README.md

metadata

inference: false
license: other

Chat & support: my new Discord server

Want to contribute? TheBloke's Patreon page

LmSys' Vicuna 33B 1.3 (final) GPTQ

These files are GPTQ 4bit model files for LmSys' Vicuna 33B 1.3 (final) merged with Kaio Ken's SuperHOT 8K.

It is the result of quantising to 4bit using GPTQ-for-LLaMa.

This is an experimental new GPTQ which offers up to 8K context size

The increased context is tested to work with ExLlama, via the latest release of text-generation-webui.

It has also been tested from Python code using AutoGPTQ, and trust_remote_code=True.

Code credits:

• Original concept and code for increasing context length: kaiokendev
• Updated Llama modelling code that includes this automatically via trust_remote_code: emozilla.

Please read carefully below to see how to use it.

NOTE: Using the full 8K context on a 30B model will exceed 24GB VRAM.

Repositories available

• 4-bit GPTQ models for GPU inference
• 2, 3, 4, 5, 6 and 8-bit GGML models for CPU inference
• Unquantised SuperHOT fp16 model in pytorch format, for GPU inference and for further conversions
• Unquantised base fp16 model in pytorch format, for GPU inference and for further conversions

How to easily download and use this model in text-generation-webui with ExLlama

Please make sure you're using the latest version of text-generation-webui

1. Click the Model tab.
2. Under Download custom model or LoRA, enter TheBloke/Vicuna-33B-1-3-SuperHOT-8K-GPTQ.
3. Click Download.
4. The model will start downloading. Once it's finished it will say "Done"
5. Untick Autoload the model
6. In the top left, click the refresh icon next to Model.
7. In the Model dropdown, choose the model you just downloaded: Vicuna-33B-1-3-SuperHOT-8K-GPTQ
8. To use the increased context, set the Loader to ExLlama, set max_seq_len to 8192 or 4096, and set compress_pos_emb to 4 for 8192 context, or to 2 for 4096 context.
9. Now click Save Settings followed by Reload
10. The model will automatically load, and is now ready for use!
11. Once you're ready, click the Text Generation tab and enter a prompt to get started!

How to use this GPTQ model from Python code with AutoGPTQ

First make sure you have AutoGPTQ and Einops installed:

pip3 install einops auto-gptq

Then run the following code. Note that in order to get this to work, config.json has been hardcoded to a sequence length of 8192.

If you want to try 4096 instead to reduce VRAM usage, please manually edit config.json to set max_position_embeddings to the value you want.

from transformers import AutoTokenizer, pipeline, logging
from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
import argparse

model_name_or_path = "TheBloke/Vicuna-33B-1-3-SuperHOT-8K-GPTQ"
model_basename = "vicuna-33b-1.3-superhot-8k-GPTQ-4bit--1g.act.order"

use_triton = False

tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True)

model = AutoGPTQForCausalLM.from_quantized(model_name_or_path,
        model_basename=model_basename,
        use_safetensors=True,
        trust_remote_code=True,
        device_map='auto',
        use_triton=use_triton,
        quantize_config=None)

model.seqlen = 8192

# Note: check the prompt template is correct for this model.
prompt = "Tell me about AI"
prompt_template=f'''USER: {prompt}
ASSISTANT:'''

print("\n\n*** Generate:")

input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda()
output = model.generate(inputs=input_ids, temperature=0.7, max_new_tokens=512)
print(tokenizer.decode(output[0]))

# Inference can also be done using transformers' pipeline

# Prevent printing spurious transformers error when using pipeline with AutoGPTQ
logging.set_verbosity(logging.CRITICAL)

print("*** Pipeline:")
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.95,
    repetition_penalty=1.15
)

print(pipe(prompt_template)[0]['generated_text'])

Using other UIs: monkey patch

Provided in the repo is llama_rope_scaled_monkey_patch.py, written by @kaiokendev.

It can be theoretically be added to any Python UI or custom code to enable the same result as trust_remote_code=True. I have not tested this, and it should be superseded by using trust_remote_code=True, but I include it for completeness and for interest.

Provided files

vicuna-33b-1.3-superhot-8k-GPTQ-4bit--1g.act.order.safetensors

This will work with AutoGPTQ, ExLlama, and CUDA versions of GPTQ-for-LLaMa. There are reports of issues with Triton mode of recent GPTQ-for-LLaMa. If you have issues, please use AutoGPTQ instead.

It was created without group_size to lower VRAM requirements, and with --act-order (desc_act) to boost inference accuracy as much as possible.

• vicuna-33b-1.3-superhot-8k-GPTQ-4bit--1g.act.order.safetensors
  • Works for use with ExLlama with increased context (4096 or 8192)
  • Works with AutoGPTQ in Python code, including with increased context, if trust_remote_code=True is set.
  • Should work with GPTQ-for-LLaMa in CUDA mode, but unknown if increased context works - TBC. May have issues with GPTQ-for-LLaMa Triton mode.
  • Works with text-generation-webui, including one-click-installers.
  • Parameters: Groupsize = -1. Act Order / desc_act = True.

Discord

For further support, and discussions on these models and AI in general, join us at:

TheBloke AI's Discord server

Thanks, and how to contribute.

Thanks to the chirper.ai team!

I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training.

If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects.

Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits.

• Patreon: https://patreon.com/TheBlokeAI
• Ko-Fi: https://ko-fi.com/TheBlokeAI

Special thanks to: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov.

Patreon special mentions: zynix , ya boyyy, Trenton Dambrowitz, Imad Khwaja, Alps Aficionado, chris gileta, John Detwiler, Willem Michiel, RoA, Mano Prime, Rainer Wilmers, Fred von Graf, Matthew Berman, Ghost , Nathan LeClaire, Iucharbius , Ai Maven, Illia Dulskyi, Joseph William Delisle, Space Cruiser, Lone Striker, Karl Bernard, Eugene Pentland, Greatston Gnanesh, Jonathan Leane, Randy H, Pierre Kircher, Willian Hasse, Stephen Murray, Alex , terasurfer , Edmond Seymore, Oscar Rangel, Luke Pendergrass, Asp the Wyvern, Junyu Yang, David Flickinger, Luke, Spiking Neurons AB, subjectnull, Pyrater, Nikolai Manek, senxiiz, Ajan Kanaga, Johann-Peter Hartmann, Artur Olbinski, Kevin Schuppel, Derek Yates, Kalila, K, Talal Aujan, Khalefa Al-Ahmad, Gabriel Puliatti, John Villwock, WelcomeToTheClub, Daniel P. Andersen, Preetika Verma, Deep Realms, Fen Risland, trip7s trip, webtim, Sean Connelly, Michael Levine, Chris McCloskey, biorpg, vamX, Viktor Bowallius, Cory Kujawski.

Thank you to all my generous patrons and donaters!

Original model card: Kaio Ken's SuperHOT 8K

SuperHOT Prototype 2 w/ 8K Context

This is a second prototype of SuperHOT, this time 30B with 8K context and no RLHF, using the same technique described in the github blog. Tests have shown that the model does indeed leverage the extended context at 8K.

You will need to use either the monkeypatch or, if you are already using the monkeypatch, change the scaling factor to 0.25 and the maximum sequence length to 8192

Looking for Merged & Quantized Models?

• 30B 4-bit CUDA: tmpupload/superhot-30b-8k-4bit-safetensors
• 30B 4-bit CUDA 128g: tmpupload/superhot-30b-8k-4bit-128g-safetensors

Training Details

I trained the LoRA with the following configuration:

• 1200 samples (~400 samples over 2048 sequence length)
• learning rate of 3e-4
• 3 epochs
• The exported modules are:
  • q_proj
  • k_proj
  • v_proj
  • o_proj
  • no bias
• Rank = 4
• Alpha = 8
• no dropout
• weight decay of 0.1
• AdamW beta1 of 0.9 and beta2 0.99, epsilon of 1e-5
• Trained on 4-bit base model

Original model card: LmSys' Vicuna 33B 1.3 (final)

Vicuna Model Card

Model Details

Vicuna is a chat assistant trained by fine-tuning LLaMA on user-shared conversations collected from ShareGPT.

• Developed by: LMSYS
• Model type: An auto-regressive language model based on the transformer architecture.
• License: Non-commercial license
• Finetuned from model: LLaMA.

Model Sources

• Repository: https://github.com/lm-sys/FastChat
• Blog: https://lmsys.org/blog/2023-03-30-vicuna/
• Paper: https://arxiv.org/abs/2306.05685
• Demo: https://chat.lmsys.org/

Uses

The primary use of Vicuna is research on large language models and chatbots. The primary intended users of the model are researchers and hobbyists in natural language processing, machine learning, and artificial intelligence.

How to Get Started with the Model

Command line interface: https://github.com/lm-sys/FastChat#vicuna-weights.
APIs (OpenAI API, Huggingface API): https://github.com/lm-sys/FastChat/tree/main#api.

Training Details

Vicuna v1.3 is fine-tuned from LLaMA with supervised instruction fine-tuning. The training data is around 140K conversations collected from ShareGPT.com. See more details in the "Training Details of Vicuna Models" section in the appendix of this paper.

Evaluation

Vicuna is evaluated with standard benchmarks, human preference, and LLM-as-a-judge. See more details in this paper.

Difference between different versions of Vicuna

See vicuna_weights_version.md