update model card
Browse files
README.md
CHANGED
@@ -1,3 +1,99 @@
|
|
1 |
---
|
2 |
license: llama2
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: llama2
|
3 |
+
language:
|
4 |
+
- it
|
5 |
+
tags:
|
6 |
+
- text-generation-inference
|
7 |
---
|
8 |
+
# Model Card for LLaMAntino-2-chat-13b-UltraChat-ITA
|
9 |
+
|
10 |
+
## Model description
|
11 |
+
|
12 |
+
<!-- Provide a quick summary of what the model is/does. -->
|
13 |
+
|
14 |
+
**LLaMAntino-2-chat-13b-UltraChat** is a *Large Language Model (LLM)* that is an instruction-tuned version of **LLaMAntino-2-chat-13b** (an italian-adapted **LLaMA 2 chat**).
|
15 |
+
This model aims to provide Italian NLP researchers with an improved model for italian dialogue use cases.
|
16 |
+
|
17 |
+
The model was trained using *QLora* and using as training data [UltraChat](https://github.com/thunlp/ultrachat) translated to the italian language using [Argos Translate](https://pypi.org/project/argostranslate/1.4.0/).
|
18 |
+
If you are interested in more details regarding the training procedure, you can find the code we used at the following link:
|
19 |
+
- **Repository:** https://github.com/swapUniba/LLaMAntino
|
20 |
+
|
21 |
+
**NOTICE**: the code has not been released yet, we apologize for the delay, it will be available asap!
|
22 |
+
|
23 |
+
- **Developed by:** Pierpaolo Basile, Elio Musacchio, Marco Polignano, Lucia Siciliani, Giuseppe Fiameni, Giovanni Semeraro
|
24 |
+
- **Funded by:** PNRR project FAIR - Future AI Research
|
25 |
+
- **Compute infrastructure:** [Leonardo](https://www.hpc.cineca.it/systems/hardware/leonardo/) supercomputer
|
26 |
+
- **Model type:** LLaMA-2-chat
|
27 |
+
- **Language(s) (NLP):** Italian
|
28 |
+
- **License:** Llama 2 Community License
|
29 |
+
- **Finetuned from model:** [swap-uniba/LLaMAntino-2-chat-13b-hf-ITA](https://huggingface.co/swap-uniba/LLaMAntino-2-chat-13b-hf-ITA)
|
30 |
+
|
31 |
+
## Prompt Format
|
32 |
+
|
33 |
+
This prompt format based on the [LLaMA 2 prompt template](https://gpus.llm-utils.org/llama-2-prompt-template/) adapted to the italian language was used:
|
34 |
+
|
35 |
+
```python
|
36 |
+
"<s>[INST] <<SYS>>\n" \
|
37 |
+
"Sei un assistente disponibile, rispettoso e onesto. " \
|
38 |
+
"Rispondi sempre nel modo piu' utile possibile, pur essendo sicuro. " \
|
39 |
+
"Le risposte non devono includere contenuti dannosi, non etici, razzisti, sessisti, tossici, pericolosi o illegali. " \
|
40 |
+
"Assicurati che le tue risposte siano socialmente imparziali e positive. " \
|
41 |
+
"Se una domanda non ha senso o non e' coerente con i fatti, spiegane il motivo invece di rispondere in modo non corretto. " \
|
42 |
+
"Se non conosci la risposta a una domanda, non condividere informazioni false.\n" \
|
43 |
+
"<</SYS>>\n\n" \
|
44 |
+
f"{user_msg_1} [/INST] {model_answer_1} </s><s>[INST] {user_msg_2} [/INST] {model_answer_2} </s> ... <s>[INST] {user_msg_N} [/INST] {model_answer_N} </s> "
|
45 |
+
```
|
46 |
+
|
47 |
+
We recommend using the same prompt in inference to obtain the best results!
|
48 |
+
|
49 |
+
## How to Get Started with the Model
|
50 |
+
|
51 |
+
Below you can find an example of model usage:
|
52 |
+
|
53 |
+
```python
|
54 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
55 |
+
|
56 |
+
model_id = "swap-uniba/LLaMAntino-2-chat-13b-hf-UltraChat-ITA"
|
57 |
+
|
58 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
59 |
+
model = AutoModelForCausalLM.from_pretrained(model_id)
|
60 |
+
|
61 |
+
user_msg = "Ciao! Come stai?"
|
62 |
+
|
63 |
+
prompt = "<s>[INST] <<SYS>>\n" \
|
64 |
+
"Sei un assistente disponibile, rispettoso e onesto. " \
|
65 |
+
"Rispondi sempre nel modo piu' utile possibile, pur essendo sicuro. " \
|
66 |
+
"Le risposte non devono includere contenuti dannosi, non etici, razzisti, sessisti, tossici, pericolosi o illegali. " \
|
67 |
+
"Assicurati che le tue risposte siano socialmente imparziali e positive. " \
|
68 |
+
"Se una domanda non ha senso o non e' coerente con i fatti, spiegane il motivo invece di rispondere in modo non corretto. " \
|
69 |
+
"Se non conosci la risposta a una domanda, non condividere informazioni false.\n" \
|
70 |
+
"<</SYS>>\n\n" \
|
71 |
+
f"{user_msg} [/INST] "
|
72 |
+
|
73 |
+
input_ids = tokenizer(prompt, return_tensors="pt").input_ids
|
74 |
+
outputs = model.generate(input_ids=input_ids, max_length=1024)
|
75 |
+
|
76 |
+
print(tokenizer.batch_decode(outputs.detach().cpu().numpy()[:, input_ids.shape[1]:], skip_special_tokens=True)[0])
|
77 |
+
```
|
78 |
+
|
79 |
+
If you are facing issues when loading the model, you can try to load it quantized:
|
80 |
+
|
81 |
+
```python
|
82 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, load_in_8bit=True)
|
83 |
+
```
|
84 |
+
|
85 |
+
*Note*: The model loading strategy above requires the [*bitsandbytes*](https://pypi.org/project/bitsandbytes/) and [*accelerate*](https://pypi.org/project/accelerate/) libraries
|
86 |
+
|
87 |
+
## Evaluation
|
88 |
+
|
89 |
+
<!-- This section describes the evaluation protocols and provides the results. -->
|
90 |
+
|
91 |
+
*Coming soon*!
|
92 |
+
|
93 |
+
## Citation
|
94 |
+
|
95 |
+
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
96 |
+
|
97 |
+
If you use this model in your research, please cite the following:
|
98 |
+
|
99 |
+
*Coming soon*!
|