wenbopan commited on
Commit
81f02fa
1 Parent(s): 5ca9d73

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +51 -5
README.md CHANGED
@@ -1,13 +1,59 @@
1
  ---
 
 
 
2
  license: mit
3
  datasets:
4
  - wenbopan/Chinese-dpo-pairs
5
  - Intel/orca_dpo_pairs
6
- language:
7
- - en
8
- - zh
9
  pipeline_tag: text-generation
10
  ---
11
- # Faro-Yi-34B-DPO
12
 
13
- Faro-Yi-34B-DPO is the DPO version of Faro-Yi-34B.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - en
4
+ - zh
5
  license: mit
6
  datasets:
7
  - wenbopan/Chinese-dpo-pairs
8
  - Intel/orca_dpo_pairs
9
+ - argilla/ultrafeedback-binarized-preferences-cleaned
10
+ - jondurbin/truthy-dpo-v0.1
 
11
  pipeline_tag: text-generation
12
  ---
 
13
 
14
+ # Faro-Yi-9B-DPO
15
+
16
+ This is the DPO version of [wenbopan/Faro-Yi-34B](https://huggingface.co/wenbopan/Faro-Yi-34B). Compared to Faro-Yi-34B and [Yi-34B-200K](https://huggingface.co/01-ai/Yi-34B-200K), the DPO model excels at many tasks, surpassing the original Yi-34B-200K by a large margin.
17
+ ## How to Use
18
+
19
+ Faro-Yi-34B-DPO uses the chatml template and performs well in both short and long contexts.
20
+
21
+
22
+ ```python
23
+ import io
24
+ import requests
25
+ from PyPDF2 import PdfReader
26
+ from vllm import LLM, SamplingParams
27
+
28
+ llm = LLM(model="wenbopan/Faro-Yi-34B-DPO", kv_cache_dtype="fp8_e5m2", max_model_len=100000)
29
+
30
+ pdf_data = io.BytesIO(requests.get("https://arxiv.org/pdf/2303.08774.pdf").content)
31
+ document = "".join(page.extract_text() for page in PdfReader(pdf_data).pages) # 100 pages
32
+
33
+ question = f"{document}\n\nAccording to the paper, what is the parameter count of GPT-4?"
34
+ messages = [ {"role": "user", "content": question} ] # 83K tokens
35
+ prompt = llm.get_tokenizer().apply_chat_template(messages, add_generation_prompt=True, tokenize=False)
36
+ output = llm.generate(prompt, SamplingParams(temperature=0.8, max_tokens=500))
37
+ print(output[0].outputs[0].text)
38
+ # Yi-9B-200K: 175B. GPT-4 has 175B \nparameters. How many models were combined to create GPT-4? Answer: 6. ...
39
+ # Faro-Yi-9B: GPT-4 does not have a publicly disclosed parameter count due to the competitive landscape and safety implications of large-scale models like GPT-4. ...
40
+ ```
41
+
42
+
43
+ <details> <summary>Or With Transformers</summary>
44
+
45
+ ```python
46
+ from transformers import AutoModelForCausalLM, AutoTokenizer
47
+
48
+ model = AutoModelForCausalLM.from_pretrained('wenbopan/Faro-Yi-34B-DPO', device_map="cuda")
49
+ tokenizer = AutoTokenizer.from_pretrained('wenbopan/Faro-Yi-34B-DPO')
50
+ messages = [
51
+ {"role": "system", "content": "You are a helpful assistant. Always answer with a short response."},
52
+ {"role": "user", "content": "Tell me what is Pythagorean theorem like you are a pirate."}
53
+ ]
54
+ input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)
55
+ generated_ids = model.generate(input_ids, max_new_tokens=512, temperature=0.5)
56
+ response = tokenizer.decode(generated_ids[0], skip_special_tokens=True) # Aye, matey! The Pythagorean theorem is a nautical rule that helps us find the length of the third side of a triangle. ...
57
+ ```
58
+
59
+ </details>