Commit
•
4c2bb6b
1
Parent(s):
70de0d5
Update README.md
Browse files
README.md
CHANGED
@@ -9,10 +9,11 @@ model-index:
|
|
9 |
|
10 |
# Llama3_8B_Odia_Unsloth
|
11 |
|
12 |
-
Llama3_8B_Odia_Unsloth is a fine-tuned Odia large language model with 8 billion parameters, and it is based on Llama3. The model is fine-tuned on
|
13 |
-
The fine-tuning uses Unsloth for faster training.
|
14 |
|
15 |
-
|
|
|
|
|
16 |
|
17 |
## Model Description
|
18 |
* Model type: A 8B fine-tuned model
|
@@ -20,6 +21,49 @@ For more details about the model, data, training procedure, and evaluations, go
|
|
20 |
* License: Llama3
|
21 |
|
22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
|
24 |
### Citation Information
|
25 |
|
@@ -38,7 +82,7 @@ If you find this model useful, please consider giving 👏 and citing:
|
|
38 |
|
39 |
### Contributions
|
40 |
|
|
|
41 |
- Sambit Sekhar
|
42 |
-
- Shantipriya Parida
|
43 |
- Debasish Dhal
|
44 |
- Shakshi Panwar
|
|
|
9 |
|
10 |
# Llama3_8B_Odia_Unsloth
|
11 |
|
12 |
+
Llama3_8B_Odia_Unsloth is a fine-tuned Odia large language model with 8 billion parameters, and it is based on Llama3. The model is fine-tuned on a comprehensive [171k Odia instruction set](https://huggingface.co/datasets/OdiaGenAI/all_combined_odia_171k), encompassing domain-specific and cultural nuances.
|
|
|
13 |
|
14 |
+
The fine-tuning process leverages Unsloth, expediting the training process for optimal efficiency.
|
15 |
+
|
16 |
+
For more details about the model, data, training procedure, and evaluations, go through the blog [post](https://www.odiagenai.org/blog/odiagenai-releases-llama3-fine-tuned-model-for-the-odia-language).
|
17 |
|
18 |
## Model Description
|
19 |
* Model type: A 8B fine-tuned model
|
|
|
21 |
* License: Llama3
|
22 |
|
23 |
|
24 |
+
## Inference
|
25 |
+
|
26 |
+
Sample inference script.
|
27 |
+
|
28 |
+
```
|
29 |
+
from unsloth import FastLanguageModel
|
30 |
+
import torch
|
31 |
+
max_seq_length = 2048
|
32 |
+
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
|
33 |
+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.
|
34 |
+
|
35 |
+
model, tokenizer = FastLanguageModel.from_pretrained(
|
36 |
+
model_name = "OdiaGenAI-LLM/Llama3_8B_Odia_Unsloth",
|
37 |
+
max_seq_length = max_seq_length,
|
38 |
+
dtype = dtype,
|
39 |
+
load_in_4bit = load_in_4bit,
|
40 |
+
)
|
41 |
+
|
42 |
+
alpaca_prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
|
43 |
+
|
44 |
+
### Instruction:
|
45 |
+
{}
|
46 |
+
|
47 |
+
### Input:
|
48 |
+
{}
|
49 |
+
|
50 |
+
### Response:
|
51 |
+
{}"""
|
52 |
+
|
53 |
+
FastLanguageModel.for_inference(model)
|
54 |
+
inputs = tokenizer(
|
55 |
+
[
|
56 |
+
alpaca_prompt.format(
|
57 |
+
"ଓଡିଶାର ରାଜଧାନୀ କ’ଣ?", # instruction
|
58 |
+
"", # input
|
59 |
+
"", # output - leave this blank for generation!
|
60 |
+
)
|
61 |
+
], return_tensors = "pt").to("cuda")
|
62 |
+
|
63 |
+
outputs = model.generate(**inputs, max_new_tokens = 512, use_cache = True)
|
64 |
+
tokenizer.batch_decode(outputs)
|
65 |
+
|
66 |
+
```
|
67 |
|
68 |
### Citation Information
|
69 |
|
|
|
82 |
|
83 |
### Contributions
|
84 |
|
85 |
+
- Dr.Shantipriya Parida
|
86 |
- Sambit Sekhar
|
|
|
87 |
- Debasish Dhal
|
88 |
- Shakshi Panwar
|