File size: 3,374 Bytes
91f1307
 
 
 
 
 
 
 
06ed7b7
 
 
 
 
6a77b68
 
7226913
6a77b68
8c5c1c4
6a77b68
1a233bb
8c5c1c4
86bb7e8
 
 
1a233bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
86bb7e8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24eec99
86bb7e8
 
 
 
 
1a233bb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7226913
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
---
license: mit
language:
- th
pipeline_tag: text-generation
tags:
- instruction-finetuning
library_name: adapter-transformers
datasets:
- iapp_wiki_qa_squad
- tatsu-lab/alpaca
- wongnai_reviews
- wisesight_sentiment
---

# 🐃🇹🇭 Buffala-LoRa-TH

Buffala-LoRA is a 7B-parameter LLaMA model finetuned to follow instructions. It is trained on the Stanford Alpaca (TH Translated), WikiTH, Pantip and IAppQ&A dataset and makes use of the Huggingface LLaMA implementation. For more information, please visit [the project's website](https://github.com/tloen/alpaca-lora).

## Issues and what next?
- The model still lacks a significant amount of world knowledge, so it is necessary to fine-tune it on larger Thai datasets > Next version: CCNet,OSCAR,thWiki
- Currently, there is no translation prompt. We plan to fine-tune the model on the SCB Thai-English dataset soon. 
- The model works well with the LangChain Search agent (Serpapi), which serves as a hotfix for world knowledge. > Plan for Spaces with search chain demo
- Lacked of chat capabilities, waiting for LangChain implementation.

## How to use

```python
import torch
from peft import PeftModel
from transformers import GenerationConfig, LlamaForCausalLM, LlamaTokenizer


device = "cuda"

tokenizer = LlamaTokenizer.from_pretrained("decapoda-research/llama-7b-hf")
model = LlamaForCausalLM.from_pretrained(
    "decapoda-research/llama-7b-hf",
    load_in_8bit=True,
    torch_dtype=torch.float16,
    device_map="auto",
)
model = PeftModel.from_pretrained(
    model,
    "Thaweewat/thai-buffala-lora-7b-v0-1",
    torch_dtype=torch.float16,
)

def generate_prompt(instruction, input=None):

    if input:
        return f"""Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Input:
{input + get_list_and_snippet(instruction)}
### Response:"""
    else:
        return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.
### Instruction:
{instruction}
### Input:
{get_list_and_snippet(instruction)}
### Response:"""

if not LOAD_8BIT:
    model.half()  

model.eval()
if torch.__version__ >= "2" and sys.platform != "win32":
    model = torch.compile(model)


def evaluate(
    instruction,
    input=None,
    temperature=0.1,
    top_p=0.75,
    top_k=40,
    num_beams=4,
    max_new_tokens=128,
    **kwargs,
):
    prompt = generate_prompt(instruction, input)
    inputs = tokenizer(prompt, return_tensors="pt")
    input_ids = inputs["input_ids"].to(device)
    generation_config = GenerationConfig(
        temperature=temperature,
        top_p=top_p,
        top_k=top_k,
        num_beams=num_beams,
        **kwargs,
    )
    with torch.no_grad():
        generation_output = model.generate(
            input_ids=input_ids,
            generation_config=generation_config,
            return_dict_in_generate=True,
            output_scores=True,
            max_new_tokens=max_new_tokens,
        )
    s = generation_output.sequences[0]
    output = tokenizer.decode(s)
    return output.split("### Response:")[1].strip()

evaluate(instruction = "จงแก้สมการต่อไปนี้ X เท่ากับเท่าไหร่", input="X+Y=15 and Y=7")
""" X = 8 """