RuterNorway/Llama-2-13b-chat-norwegian-GPTQ

Llama 2 13b Chat Norwegian GPTQ

This is a GPTQ version of Llama 2 13b Norwegian

Read more about GPTQ here. GPTQ models are less computationally intensive to run, but their performance accuracy is often significantly lower than full models. In testing, this model has demonstrated its ability to reply in Norwegian; however, its knowledge and quality of responses are limited.

For a demo script, see here.

Llama-2-13b-chat-norwegian-GPTQ is a variant of Meta´s Llama 2 13b Chat model, finetuned on a mix of norwegian datasets created in Ruter AI Lab the summer of 2023.

The model is tuned to understand and generate text in Norwegian. It's trained for one epoch on norwegian-alpaca + 15000 samples of machine-translated data from OpenOrca. A small subset of custom-made instructional data is also included.

For other versions of this model see:

• Llama-2-13b-chat-norwegian
• Llama-2-13b-chat-norwegian-LoRa
• Llama-2-13b-chat-norwegian-GPTQ

Data

• Norwegian alpaca
• 15k Norwegian OpenOrcra (to be released)
• Small subset of custom made instructional data

Intended Use

This model is intended for commercial and research use in Norwegian and can be used as an assistant-like chat.

Prompt Template

Llama2 Chat uses a new prompt format:

<s>[INST] <<SYS>>
You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information. Please answer in the same language as the user.
<</SYS>>
This is a test question[/INST] This is a answer </s><s>

See also the original implementation here.

We also implemented the alpaca prompt format, which the model supports.:

### Instruction:
Summarize following text.
### Input:
Text to be summarized
### Response:

Why this model?

As a Norwegian company, we understand firsthand the pressing need for powerful language models tailored to specific languages. Our primary focus is on the Norwegian linguistic landscape. In the age of digitization, languages that lack robust, open-source models can risk becoming marginalized. This is why we're introducing this open-source Norwegian model. We believe that by making such resources freely accessible, we can democratize information, foster innovation, and create a more inclusive digital ecosystem. Our aspiration is for this model to serve as a foundational resource for future specialized Norwegian models. Ultimately, our goal is to bolster the Norwegian NLP community and facilitate the smoother integration of Norwegian models into diverse projects.

Limitations

• This is an LLM, not a knowledge model. It can not be expected to have more information about Norway than the basemodel.
• It will generally preform better on tasks that involves summarization, question answering and chat, than on tasks that requires more knowledge about Norway, specific domains, or tasks where the model can answer freely.
• The data used for training is machine translated, and may contain grammatical errors and other errors.
• The model is released as is, and would in most cases need prompt tuning to achieve optimal results.

License

Llama 2 is licensed under the LLAMA 2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. See the original model card for more information.

From norwegian-alpaca we also note that "the current version uses OpenAI's gpt-3.5-turbo; hence, this dataset cannot be used to create models that compete in any way against OpenAI."

Disclaimer

• The model is available "as is". Ruter As takes no responsibility for further use.
• During testing, it seems that the safeguards implemented by Meta, still work as expected in this model. However, we want to point to the Ethical Considerations and Limitations from the origenal model card:

Llama 2 is a new technology that carries risks with use. Testing conducted to date has been in English, and has not covered, nor could it cover all scenarios.
For these reasons, as with all LLMs, Llama 2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts.
Therefore, before deploying any applications of Llama 2, developers should perform safety testing and tuning tailored to their specific applications of the model.
Please see the Responsible Use Guide available at https://ai.meta.com/llama/responsible-use-guide/

Credits

This model was made at Ruters AI Lab - summer of 2023 as part of their AI initiative.

The team wants to thank the support we got from the entire Ruter organization, and especially the Data Science team.

---

Llama 2 13b Chat Norwegian GPTQ (Norsk)

Dette er en GPTQ versjon av Llama 2 13b Chat Norwegian modellen Llama-2-13b-chat-norwegian-GPQT er en versjon av Meta sin Llama 2 13b Chat model, finetuned på en kombinasjon av diverse norske datasett. Modellen ble laget i Ruter AI Lab 2023.

Les mer om GPTQ her.

For demo script, se her.

Andre versjoner av modellen:

• Llama-2-13b-chat-norwegian
• Llama-2-13b-chat-norwegian-LoRa
• Llama-2-13b-chat-norwegian-GPTQ

Modellen er finetuned til å forstå og generere tekst på Norsk. Den er trent i én epoch med norwegian-alpaca + et utvalg av 15000 maskinoversatt data fra OpenOrca. Det består og av et lite sett med selvlagde instruksjonsdata.

This is a GPTQ version of Llama 2 13b Norwegian

Llama-2-13b-chat-norwegian is a variant of Meta´s Llama 2 13b Chat model, finetuned on a mix of norwegian datasets created in Ruter AI Lab the summer of 2023.

The model is tuned to understand and generate text in Norwegian. It's trained for one epoch on norwegian-alpaca + 15000 samples of machine-translated data from OpenOrca (the dataset to be released). A small subset of custom-made instructional data is also included.

Data

• Norwegian alpaca
• 15k Norwegian OpenOrcra (venter på utgivelse)
• Lite sett med selvlagde instruksjonsdata

Prompt Mal

Llama2 Chat bruker et nytt prompt format:

<s>[INST] <<SYS>>
You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.
If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information. Please answer in the same language as the user.
<</SYS>>
This is a test question[/INST] This is a answer </s><s>

Se orgianl implementasjon her.

Vi har også implementert alpaca prompt formatet, som også er støttet av modellen.

### Instruction:
Summarize following text.
### Input:
Text to be summarized
### Response:

Hvorfor denne modellen?

Som et norsk selskap forstår vi selv det presserende behovet for kraftige språkmodeller tilpasset spesifikke språk. Vårt primære fokus er på det norske språkområdet. I den digitale alderen risikerer språk som mangler robuste, åpne kildekodemodeller å bli marginalisert. Dette er grunnen til at vi nå introduserer denne åpne kildekodemodellen for norsk. Vi tror at ved å gjøre disse ressursene tilgjengelige gratis, kan vi demokratisere informasjonen, fremme innovasjon og skape et mer inkluderende digitalt økosystem. Vår ambisjon er at denne modellen skal tjene som en grunnleggende ressurs for fremtidige spesialiserte norske modeller. Vårt mål er å styrke det norske NLP-miljøet og gjøre det enklere å innlemme norske modeller i ulike prosjekter.

Begrensninger

• Dette er en LLM, ikke en kunnskapsmodell. Den kan ikke forventes å ha mer informasjon om Norge enn basismodellen.
• Den vil generelt prestere bedre på oppgaver som innebærer oppsummering, spørsmålsbesvarelse og chat, enn på oppgaver som krever mer kunnskap om Norge, spesifikke domener, eller oppgaver hvor modellen kan svare fritt.
• Dataene som brukes til trening er maskinoversatt, og kan inneholde grammatiske feil. Vi har kun gjort en rask manuell sjekk av dataene.
• Modellen er utgitt som den er, og vil i de fleste tilfeller trenge "prompt tuning" for å oppnå ønskede resultater.

Lisens

Llama 2 er lisensiert under LLAMA 2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Se det orginale modell kortet for mer informasjon.

Fra norwegian-alpaca vil vi gjøre oppmerksomme på at "the current version uses OpenAI's gpt-3.5-turbo; hence, this dataset cannot be used to create models that compete in any way against OpenAI."

Ansvarsfraskrivelse

• Modellen tilgjengeliggjøres «som den er». Ruter As tar ikke noe ansvar for videre bruk.
• Under testingen virket det som sikkerhetstiltakene implementert av Meta fortsatt fungerer som forventet for denne modellen. Vi gjør derimot oppmerksom på de etiske betraktiningene og begrensningene fra det orignale modellkortet:

Llama 2 is a new technology that carries risks with use. Testing conducted to date has been in English, and has not covered, nor could it cover all scenarios.
For these reasons, as with all LLMs, Llama 2’s potential outputs cannot be predicted in advance, and the model may in some instances produce inaccurate, biased or other objectionable responses to user prompts.
Therefore, before deploying any applications of Llama 2, developers should perform safety testing and tuning tailored to their specific applications of the model.
Please see the Responsible Use Guide available at https://ai.meta.com/llama/responsible-use-guide/

Demo script

Google Colab implementation

How to use this GPTQ model from Python code

To run the GPTQ version you first need to install transformers, peft, accelerate optimum and auto-gptq

pip install -q -U transformers peft accelerate optimum
pip install auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu117/

Then you can run the model with:

# Loading the model

from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline

model_id = "RuterNorway/Llama-2-13b-chat-norwegian-GPTQ"
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Regular prompting
instruction = "Oppsumer følgende tekst kort"
input = "Ruter As er et kollektivtrafikkselskap dannet 23. oktober 2007 ved fusjon mellom Stor-Oslo Lokaltrafikk (SL) og AS Oslo Sporveier.\nFunksjon\nSelskapet er en såkalt bestillerenhet for kollektivtrafikken i Oslo og deler av Viken, i hovedsak Akershus-delen. Det geografiske dekningsområdet til selskapet er inndelt i ulike soner, se sonekart. Den 1. januar 2008 overtok selskapet trafikkdelen av virksomheten til AS Oslo Sporveier (OS) og Stor-Oslo Lokaltrafikk AS (SL). Selskapet er eid av Oslo kommune (60 prosent) og Viken fylkeskommune (40 prosent). Selskapet har som oppgave å administrere og planlegge all kollektivtrafikk i Oslo og deler av Viken, bortsett fra Vys togtrafikk. Ruter AS har rundt 90 ansatte.\nFor de reisende medførte opprettelsen av det nye selskapet i første omgang lite, men reisende med enkeltbillett kan bruke billetten til overgang til andre selskaper innenfor Ruters område. Overgangen til et mer fleksibelt rutetilbud med bussmating til skinnegående transport gav etter hvert vekst for selskapet med betydelig overgang fra bil til kollektivtrafikk.\nOmfang \nDet ble foretatt 387 millioner reiser med kollektivtransport i Oslo og Ruters del av Viken i 2018. Kollektivtrafikken i Ruters trafikkområde tilsvarer om lag halvparten av all kollektivtrafikk i Norge. \nUtslippsfri drift\n Selskapet har som ambisjon at alle busser, båter, minibusser og drosjer som kjører for Ruter skal ha utslippsfri drift i løpet av 2028. Trikk og t-bane går allerede på strøm. Selskapet har beregnet at spesialtransporten kan være utslippsfri i 2022, båtene i 2024 og alle bussene i 2028."
# Text from https://snl.no/Ruter_As
prompt_template=f'''### Instruction: {instruction}
### Input: {input}
### Response:
'''
print("\n\n*** Generate:")
inputs = tokenizer(prompt_template, return_tensors="pt").to(0)

out = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(out[0], skip_special_tokens=True))

# Pipeline prompting
print("\n\n*** Pipeline:\n\n")
pipe = pipeline(
    "text-generation",
    model=model,
    do_sample=True,
    tokenizer=tokenizer,
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.95,
    repetition_penalty=1.15
)
print(pipe(prompt_template)[0]['generated_text'][len(prompt_template):])

Compatibility

The files provided will work with AutoGPTQ (CUDA and Triton modes), GPTQ-for-LLaMa (only CUDA has been tested), ExLlama, and Occ4m's GPTQ-for-LLaMa fork.