Update README.md
Browse files
README.md
CHANGED
|
@@ -29,6 +29,79 @@ This approach ensures:
|
|
| 29 |
|
| 30 |
---
|
| 31 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
## 📊 GCPO Improves Reasoning Performance
|
| 33 |
|
| 34 |
GCPO consistently outperforms DAPO.
|
|
|
|
| 29 |
|
| 30 |
---
|
| 31 |
|
| 32 |
+
## 🛠️ Model Use
|
| 33 |
+
|
| 34 |
+
### ✅ Use with Hugging Face Transformers
|
| 35 |
+
|
| 36 |
+
```python
|
| 37 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 38 |
+
|
| 39 |
+
model_name = "Ach0/GCPO-R1-1.5B"
|
| 40 |
+
|
| 41 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
| 42 |
+
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto", trust_remote_code=True)
|
| 43 |
+
|
| 44 |
+
question = """
|
| 45 |
+
Solve the following math problem efficiently and clearly. The last line of your response should be of the following format: 'Therefore, the final answer is: $\\boxed{{ANSWER}}$. I hope it is correct' (without quotes) where ANSWER is just the final number or expression that solves the problem. Think step by step before answering.
|
| 46 |
+
|
| 47 |
+
Point $B$ is on $\\overline{AC}$ with $AB = 9$ and $BC = 21.$ Point $D$ is not on $\\overline{AC}$ so that $AD = CD,$ and $AD$ and $BD$ are integers. Let $s$ be the sum of all possible perimeters of $\\triangle ACD$. Find $s.$
|
| 48 |
+
"""
|
| 49 |
+
|
| 50 |
+
messages = [
|
| 51 |
+
{"role": "user", "content": question}
|
| 52 |
+
]
|
| 53 |
+
|
| 54 |
+
prompt = tokenizer.apply_chat_template(
|
| 55 |
+
messages,
|
| 56 |
+
tokenize=False,
|
| 57 |
+
add_generation_prompt=True,
|
| 58 |
+
)
|
| 59 |
+
|
| 60 |
+
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
|
| 61 |
+
outputs = model.generate(**inputs, max_new_tokens=8192)
|
| 62 |
+
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
|
| 63 |
+
```
|
| 64 |
+
|
| 65 |
+
### ✅ Use with vLLM(fast inference)
|
| 66 |
+
|
| 67 |
+
```python
|
| 68 |
+
from vllm import LLM, SamplingParams
|
| 69 |
+
from transformers import AutoTokenizer
|
| 70 |
+
|
| 71 |
+
model_name = "Ach0/GCPO-R1-1.5B"
|
| 72 |
+
|
| 73 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
| 74 |
+
llm = LLM(model=model_name, trust_remote_code=True)
|
| 75 |
+
|
| 76 |
+
sampling_params = SamplingParams(
|
| 77 |
+
temperature=0.7,
|
| 78 |
+
top_p=0.8,
|
| 79 |
+
top_k=20,
|
| 80 |
+
max_tokens=8192
|
| 81 |
+
)
|
| 82 |
+
|
| 83 |
+
question = """
|
| 84 |
+
Solve the following math problem efficiently and clearly. The last line of your response should be of the following format: 'Therefore, the final answer is: $\\boxed{{ANSWER}}$. I hope it is correct' (without quotes) where ANSWER is just the final number or expression that solves the problem. Think step by step before answering.
|
| 85 |
+
|
| 86 |
+
Point $B$ is on $\\overline{AC}$ with $AB = 9$ and $BC = 21.$ Point $D$ is not on $\\overline{AC}$ so that $AD = CD,$ and $AD$ and $BD$ are integers. Let $s$ be the sum of all possible perimeters of $\\triangle ACD$. Find $s.$
|
| 87 |
+
"""
|
| 88 |
+
|
| 89 |
+
messages = [
|
| 90 |
+
{"role": "user", "content": question}
|
| 91 |
+
]
|
| 92 |
+
|
| 93 |
+
prompt = tokenizer.apply_chat_template(
|
| 94 |
+
messages,
|
| 95 |
+
tokenize=False,
|
| 96 |
+
add_generation_prompt=True
|
| 97 |
+
)
|
| 98 |
+
|
| 99 |
+
outputs = llm.generate([prompt], sampling_params)
|
| 100 |
+
print(outputs[0].outputs[0].text)
|
| 101 |
+
```
|
| 102 |
+
|
| 103 |
+
---
|
| 104 |
+
|
| 105 |
## 📊 GCPO Improves Reasoning Performance
|
| 106 |
|
| 107 |
GCPO consistently outperforms DAPO.
|