Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# internlm2-chat-20b-llama
|
2 |
+
|
3 |
+
[`internlm/internlm2-20b`](https://huggingface.co/internlm/internlm2-20b) weights are formatted to match standard Llama modeling code.
|
4 |
+
Model can be loaded directly, but for tokenizer use `trust_remote_code`
|
5 |
+
|
6 |
+
# usage:
|
7 |
+
```py
|
8 |
+
import torch
|
9 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
10 |
+
|
11 |
+
model_name = "kiranr/internlm2-chat-20b-llama"
|
12 |
+
|
13 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
14 |
+
|
15 |
+
model = AutoModelForCausalLM.from_pretrained(
|
16 |
+
model_name,
|
17 |
+
torch_dtype=torch.float16,
|
18 |
+
device_map="auto",
|
19 |
+
attn_implementation="flash_attention_2",
|
20 |
+
)
|
21 |
+
messages = [
|
22 |
+
{"role": "user", "content": "what is the square root of banana?"}
|
23 |
+
]
|
24 |
+
|
25 |
+
model_input = tokenizer.apply_chat_template(messages, return_tensors="pt").to("cuda")
|
26 |
+
|
27 |
+
generated_ids = model.generate(
|
28 |
+
model_input,
|
29 |
+
max_new_tokens=1024,
|
30 |
+
do_sample=True,
|
31 |
+
eos_token_id=[92542, 2], # <|im_end|> and </s>
|
32 |
+
)
|
33 |
+
output = tokenizer.decode(
|
34 |
+
generated_ids[0][model_input.shape[-1] : -1], skip_special_tokens=True
|
35 |
+
)
|
36 |
+
print(output)
|
37 |
+
```
|