RichardErkhov/tartuNLP_-_Llammas-gguf

Quantization made by Richard Erkhov.

Github

Discord

Request more models

Llammas - GGUF

• Model creator: https://huggingface.co/tartuNLP/
• Original model: https://huggingface.co/tartuNLP/Llammas/

Name	Quant method	Size
Llammas.Q2_K.gguf	Q2_K	2.36GB
Llammas.IQ3_XS.gguf	IQ3_XS	2.6GB
Llammas.IQ3_S.gguf	IQ3_S	2.75GB
Llammas.Q3_K_S.gguf	Q3_K_S	2.75GB
Llammas.IQ3_M.gguf	IQ3_M	2.9GB
Llammas.Q3_K.gguf	Q3_K	3.07GB
Llammas.Q3_K_M.gguf	Q3_K_M	3.07GB
Llammas.Q3_K_L.gguf	Q3_K_L	3.35GB
Llammas.IQ4_XS.gguf	IQ4_XS	3.4GB
Llammas.Q4_0.gguf	Q4_0	3.56GB
Llammas.IQ4_NL.gguf	IQ4_NL	3.58GB
Llammas.Q4_K_S.gguf	Q4_K_S	3.59GB
Llammas.Q4_K.gguf	Q4_K	3.8GB
Llammas.Q4_K_M.gguf	Q4_K_M	3.8GB
Llammas.Q4_1.gguf	Q4_1	3.95GB
Llammas.Q5_0.gguf	Q5_0	4.33GB
Llammas.Q5_K_S.gguf	Q5_K_S	4.33GB
Llammas.Q5_K.gguf	Q5_K	4.45GB
Llammas.Q5_K_M.gguf	Q5_K_M	4.45GB
Llammas.Q5_1.gguf	Q5_1	4.72GB
Llammas.Q6_K.gguf	Q6_K	5.15GB
Llammas.Q8_0.gguf	Q8_0	6.67GB

Original model description:

language: - et - en pipeline_tag: text-generation library_name: transformers tags: - conversational

LLammas 🐑

Llama-2-7B finetuned in two stages:

1. 5B tokens of CulturaX with 75% of documents in Estonain and 25% in English (see Llammas-base),
2. Alpaca-cleaned, Alpaca-est, OASST1 top-1 English conversations, CoT and FLAN-V2 following open-instruct (both 10,000), WMT18 English-Estonian translation development data (as documents), general MTee validation English-Estonian held-out data.

Alpaca-est is an instruction dataset generated for Estonian with gpt-3.5-turbo-0613, following Alpaca. More details in our paper.

Additional resources:

• Paper: arxiv.org/abs/2404.04042
• Code: github.com/TartuNLP/llammas
• Base model: tartuNLP/Llammas-base
• 4-bit quantized model in GGUF: AlbertUnn/LlammasGGUF
• Alpaca-est dataset: github.com/TartuNLP/alpaca-est

Using the model

Using the model in a text-generation pipeline:

from transformers import pipeline
import torch

pipe = pipeline("text-generation", model="tartuNLP/Llammas", torch_dtype=torch.bfloat16, device_map="auto")

messages = [
    {"role": "user", "content": "Tere!"},
    {"role": "assistant", "content": "Tere! Kas saaksin teid kuidagi aidata?"},
    {"role": "user", "content": "Kuidas alustada kirja kirjutamist?"}
]

prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.6, top_k=50, top_p=0.9)
print(outputs[0]["generated_text"][len(prompt):])

Using the model in a conversational pipeline (works with transformers==4.36.2, issues with output in newer versions):

from transformers import pipeline, Conversation
import torch

pipe = pipeline("conversational", model="tartuNLP/Llammas", torch_dtype=torch.bfloat16, device_map="auto")

messages = [
    {"role": "user", "content": "Tere!"},
    {"role": "assistant", "content": "Tere! Kas saaksin teid kuidagi aidata?"},
    {"role": "user", "content": "Kuidas alustada kirja kirjutamist?"}
]

conversation = Conversation(messages)
conversation = pipe(conversation)

Conversational format:

<|user|>
Tere!
<|assistant|>
Tere! Kas saaksin teid kuidagi aidata?</s>
<|user|>
Kuidas alustada kirja kirjutamist?
<|assistant|>
Kirja kirjutamiseks alustage tervitusega, näiteks "Tere!" või "Tere hommikust!". Seejärel tutvustage ennast ja mainige, kellega kirjutate. Kirjeldage oma mõtteid või küsimusi, mida soovite arutada. Lõpetage kiri viisakalt, näiteks "Tänan teid tähelepanu eest!" või "Parimate soovidega!"</s>

Citation

@misc{kuulmets2024teaching,
      title={Teaching Llama a New Language Through Cross-Lingual Knowledge Transfer}, 
      author={Hele-Andra Kuulmets and Taido Purason and Agnes Luhtaru and Mark Fishel},
      year={2024},
      eprint={2404.04042},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}