TimeMobius
commited on
Commit
•
f6764e5
1
Parent(s):
14b499e
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,64 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
+
language:
|
4 |
+
- fr
|
5 |
+
- it
|
6 |
+
- de
|
7 |
+
- es
|
8 |
+
- en
|
9 |
+
- zh
|
10 |
+
inference: false
|
11 |
---
|
12 |
+
# Model Card for Mobius-12B-base-m1
|
13 |
+
The Mobius-12B-base-m1 Large Language Model (LLM) is a pretrained model based on RWKV v5 arch. We use
|
14 |
+
|
15 |
+
|
16 |
+
## Warning
|
17 |
+
This repo contains weights that are not compatible with Hugging Face [transformers](https://github.com/huggingface/transformers) library yet. But you can try this[PR]() as well.
|
18 |
+
[RWKV runner]() or [AI00 server]() also work.
|
19 |
+
|
20 |
+
## Instruction|Chat format
|
21 |
+
|
22 |
+
This format must be strictly respected, otherwise the model will generate sub-optimal outputs.
|
23 |
+
|
24 |
+
The template used to build a prompt for the Instruct model is defined as follows:
|
25 |
+
```
|
26 |
+
User: {Instruction|prompt}\n\nAssistant:
|
27 |
+
```
|
28 |
+
|
29 |
+
## Run the model
|
30 |
+
|
31 |
+
```python
|
32 |
+
import torch
|
33 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
34 |
+
|
35 |
+
model = AutoModelForCausalLM.from_pretrained("TimeMobius/Mobius-12B-base-m1", torch_dtype=torch.float16).to(0)
|
36 |
+
tokenizer = AutoTokenizer.from_pretrained("TimeMobius/Mobius-12B-base-m1", trust_remote_code=True)
|
37 |
+
|
38 |
+
text = "x"
|
39 |
+
prompt = f'Question: {text.strip()}\n\nAnswer:'
|
40 |
+
|
41 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(0)
|
42 |
+
output = model.generate(inputs["input_ids"], max_new_tokens=40)
|
43 |
+
print(tokenizer.decode(output[0].tolist(), skip_special_tokens=True))
|
44 |
+
```
|
45 |
+
|
46 |
+
## Limitations
|
47 |
+
|
48 |
+
The Mobius base m1 is the base model can be easily fine-tuned to achieve compelling performance.
|
49 |
+
|
50 |
+
### Benchmark
|
51 |
+
|
52 |
+
| Mobius-12B-base-m1 | |
|
53 |
+
|--------------------|----------|
|
54 |
+
| ppl | 3.41 |
|
55 |
+
| piqa | 0.78 |
|
56 |
+
| hellaswag | 0.71 |
|
57 |
+
| winogrande | 0.68 |
|
58 |
+
| arc_challenge | 0.42 |
|
59 |
+
| arc_easy | 0.73 |
|
60 |
+
| openbookqa | 0.40 |
|
61 |
+
| sciq | 0.93 |
|
62 |
+
|
63 |
+
|
64 |
+
# @TimeMobius
|