You can deploy this model and start earning money today!

RichardErkhov/
mesolitica_-_malaysian-mistral-7b-32k-instructions-v4-gguf

Name: mesolitica_-_malaysian-mistral-7b-32k-instructions-v4-gguf
Availability: InStock

Views

No views yet

Quantization made by Richard Erkhov.

Github

Discord

Request more models

malaysian-mistral-7b-32k-instructions-v4 - GGUF

Model creator: https://huggingface.co/mesolitica/
Original model: https://huggingface.co/mesolitica/malaysian-mistral-7b-32k-instructions-v4/

Name	Quant method	Size
malaysian-mistral-7b-32k-instructions-v4.Q2_K.gguf	Q2_K	2.53GB
malaysian-mistral-7b-32k-instructions-v4.Q3_K_S.gguf	Q3_K_S	2.95GB
malaysian-mistral-7b-32k-instructions-v4.Q3_K.gguf	Q3_K	3.28GB
malaysian-mistral-7b-32k-instructions-v4.Q3_K_M.gguf	Q3_K_M	3.28GB
malaysian-mistral-7b-32k-instructions-v4.Q3_K_L.gguf	Q3_K_L	3.56GB
malaysian-mistral-7b-32k-instructions-v4.IQ4_XS.gguf	IQ4_XS	3.67GB
malaysian-mistral-7b-32k-instructions-v4.Q4_0.gguf	Q4_0	3.83GB
malaysian-mistral-7b-32k-instructions-v4.IQ4_NL.gguf	IQ4_NL	3.87GB
malaysian-mistral-7b-32k-instructions-v4.Q4_K_S.gguf	Q4_K_S	3.86GB
malaysian-mistral-7b-32k-instructions-v4.Q4_K.gguf	Q4_K	4.07GB
malaysian-mistral-7b-32k-instructions-v4.Q4_K_M.gguf	Q4_K_M	4.07GB
malaysian-mistral-7b-32k-instructions-v4.Q4_1.gguf	Q4_1	4.24GB
malaysian-mistral-7b-32k-instructions-v4.Q5_0.gguf	Q5_0	4.65GB
malaysian-mistral-7b-32k-instructions-v4.Q5_K_S.gguf	Q5_K_S	4.65GB
malaysian-mistral-7b-32k-instructions-v4.Q5_K.gguf	Q5_K	4.78GB
malaysian-mistral-7b-32k-instructions-v4.Q5_K_M.gguf	Q5_K_M	4.78GB
malaysian-mistral-7b-32k-instructions-v4.Q5_1.gguf	Q5_1	5.07GB
malaysian-mistral-7b-32k-instructions-v4.Q6_K.gguf	Q6_K	5.53GB
malaysian-mistral-7b-32k-instructions-v4.Q8_0.gguf	Q8_0	7.17GB

Original model description:

language:

Full Parameter Finetuning 7B 32768 context length Mistral on Malaysian instructions dataset

README at https://github.com/mesolitica/malaya/tree/5.1/session/mistral#instructions-7b-16384-context-length

We use exact Mistral Instruct chat template.

WandB, https://wandb.ai/huseinzol05/fpf-mistral-7b-hf-instructions-16k?workspace=user-huseinzol05

WandB report, https://wandb.ai/huseinzol05/fpf-tinyllama-1.1b-hf-instructions-16k/reports/Instruction-finetuning--Vmlldzo2MzQ3OTcz

Dataset

Dataset gathered at https://huggingface.co/collections/mesolitica/malaysian-synthetic-dataset-656c2673fe7fe0b1e9e25fe2

Notebook to prepare dataset at https://github.com/mesolitica/malaysian-dataset/blob/master/llm-instruction/combine-malay-no-alignment-multitasks-v5.ipynb

Limitations

This model is a quick demonstration that the base model can be easily fine-tuned to achieve some performance. It does have minimal moderation mechanisms.

how-to

python

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

TORCH_DTYPE = 'bfloat16'
nf4_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type='nf4',
    bnb_4bit_use_double_quant=True,
    bnb_4bit_compute_dtype=getattr(torch, TORCH_DTYPE)
)

tokenizer = AutoTokenizer.from_pretrained('mesolitica/malaysian-mistral-7b-32k-instructions-v4')
model = AutoModelForCausalLM.from_pretrained(
    'mesolitica/malaysian-mistral-7b-32k-instructions-v4',
    use_flash_attention_2 = True,
    quantization_config = nf4_config
)

messages = [
    {'role': 'user', 'content': 'kwsp tu apa'}
]
prompt = tokenizer.apply_chat_template(messages, tokenize = False)
inputs = tokenizer([prompt], return_tensors='pt', add_special_tokens=False).to('cuda')
generate_kwargs = dict(
    inputs,
    max_new_tokens=1024,
    top_p=0.95,
    top_k=50,
    temperature=0.9,
    do_sample=True,
    num_beams=1,
)
r = model.generate(**generate_kwargs)
tokenizer.decode(r[0])

<s> [INST] kwsp tu apa [/INST]KWSP bermaksud Kumpulan Wang Simpanan Pekerja. Ia adalah sebuah institusi simpanan persaraan yang ditubuhkan oleh Kementerian Kewangan Malaysia untuk tujuan mengumpul simpanan ahli untuk dibayar pada umur persaraan, penuh atau penuh persaraan penuh. KWSP ditubuhkan pada tahun 1951 dan mula beroperasi pada tahun 1952. KWSP adalah salah satu institusi simpanan persaraan terbesar di dunia, dengan pangkalan ahli sekitar 14 juta ahli.</s>

RichardErkhov/mesolitica_-_malaysian-mistral-7b-32k-instructions-v4-gguf