Improve model card: update pipeline tag, add library name, and fix tags

README.md

@@ -1,21 +1,22 @@
 ---
-license: cc-by-nc-sa-4.0
+base_model:
+- Qwen/Qwen2.5-7B-Instruct
 language:
 - en
 - zh
-base_model:
-- Qwen/Qwen2.5-7B-Instruct
+license: cc-by-nc-sa-4.0
+pipeline_tag: translation
+library_name: transformers
 tags:
-- machine tranlsation
+- machine translation
 - O1-like model
 - Chat
-pipeline_tag: text-generation
 ---
 
 # DRT
 
 <p align="center">
-🤗 <a href="https://huggingface.co/Krystalan/DRT-7B">DRT-7B</a>&nbsp&nbsp | &nbsp&nbsp🤗 <a href="https://huggingface.co/Krystalan/DRT-8B">DRT-8B</a>&nbsp&nbsp | &nbsp&nbsp🤗 <a href="https://huggingface.co/Krystalan/DRT-14B">DRT-14B</a>&nbsp&nbsp | &nbsp&nbsp 📑 <a href="https://arxiv.org/abs/2412.17498">Paper</a>
+🤗 <a href="https://huggingface.co/Krystalan/DRT-7B">DRT-7B</a>&nbsp&nbsp | &nbsp&nbsp🤗 <a href="https://huggingface.co/Krystalan/DRT-8B">DRT-8B</a>&nbsp&nbsp | &nbsp&nbsp🤗 <a href="https://huggingface.co/Krystalan/DRT-14B">DRT-14B</a>&nbsp&nbsp | &nbsp&nbsp 📑 <a href="https://arxiv.org/abs/2412.17498">Paper</a>&nbsp&nbsp | &nbsp&nbsp 💻 <a href="https://github.com/krystalan/DRT-o1">Code</a>
 
 </p>
 
@@ -80,7 +81,8 @@ In this work, we introduce DRT, an attempt to bring the success of long thought
 ### Model Prompts
 During model inference, please use the following prompts:
 - System prompt: `You are a philosopher skilled in deep thinking, accustomed to exploring complex problems with profound insight.`
-- User prompt: `Please translate the following text from English to Chinese:\n[An English text]`
+- User prompt: `Please translate the following text from English to Chinese:
+[An English text]`
 
 DRT models will first generate the thought and then provide the final translation, with the following format:
 ```
@@ -107,7 +109,8 @@ model = AutoModelForCausalLM.from_pretrained(
 )
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 
-prompt = "Please translate the following text from English to Chinese:\nThe mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap."
+prompt = "Please translate the following text from English to Chinese:
+The mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap."
 messages = [
     {"role": "system", "content": "You are a philosopher skilled in deep thinking, accustomed to exploring complex problems with profound insight."},
     {"role": "user", "content": prompt}
@@ -154,7 +157,8 @@ chat_response = client.chat.completions.create(
     model=[model_name],
     messages=[
         {"role": "system", "content": "You are a philosopher skilled in deep thinking, accustomed to exploring complex problems with profound insight."},
-        {"role": "user", "content": "Please translate the following text from English to Chinese:\nThe mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap."},
+        {"role": "user", "content": "Please translate the following text from English to Chinese:
+The mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap."},
     ],
     temperature=0.1,
     top_p=0.8,
@@ -177,8 +181,222 @@ print("Chat response:", chat_response)
 
 
 
-## License
-This work is licensed under cc-by-nc-sa-4.0
+## Data
+
+We release the synthesized data (named ```MetaphorTrans```), please refer to `data/MetaphorTrans_*.jsonl`, where `text` and `trans` denote the source English sentences and the target Chinese translations, respectively. `thought` indicates the thought content for MT.
+
+
+# DeepTrans
+
+![](./images/deeptrans-reward-framework.png)
+
+In this work, we propose DeepTrans-7B, which aims at enhancing the free translation ability of deep reasoning LLMs via RL. To this end, we use DeepSeek-v3 (671B) as the reward model, and design scoring criteria on both translations and thought process.
+
+## Model Checkpoint
+
+|  | Backbone | Model Access |
+| :--: | :--: | :--: |
+| DeepTrans-7B | 🤗 <a href="https://huggingface.co/Qwen/Qwen2.5-7B-Instruct">Qwen2.5-7B-Instruct</a> | 🤗 <a href="https://huggingface.co/Krystalan/DeepTrans-7B">DeepTrans-7B</a> |
+
+## Inference
+
+- Huggingface Transformers
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model_name = "Krystalan/DeepTrans-7B"
+
+model = AutoModelForCausalLM.from_pretrained(
+    model_name,
+    torch_dtype="auto",
+    device_map="auto"
+)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+prompt = "你是一个翻译专家，擅长将英文翻译成中文。你在翻译过程中非常擅长思考，会先进行思考再给出翻译结果。你的输出格式为：
+<think>
+[思考过程]
+</think>[翻译结果]
+
+在你思考完之后，也就是</think>之后，你会给出最终的翻译即“[翻译结果]”，且[翻译结果]中不需要给出任何解释和描述，只需要提供英文的翻译结果。
+现在请你翻译以下这句英语：
+" + "The mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap."
+
+messages = [
+    {"role": "user", "content": prompt}
+]
+text = tokenizer.apply_chat_template(
+    messages,
+    tokenize=False,
+    add_generation_prompt=True
+)
+model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
+
+generated_ids = model.generate(
+    **model_inputs,
+    max_new_tokens=2048
+)
+generated_ids = [
+    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+]
+
+response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+print(response)
+```
+
+- VLLM:
+deploying LLMs:
+```bash
+python3 -m vllm.entrypoints.openai.api_server --model [model_ckpt] --served-model-name [model_name]
+```
+
+calling LLMs:
+```python
+from openai import OpenAI
+# Set OpenAI's API key and API base to use vLLM's API server.
+openai_api_key = "EMPTY"
+openai_api_base = "http://localhost:8000/v1"
+
+client = OpenAI(
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+prompt = "你是一个翻译专家，擅长将英文翻译成中文。你在翻译过程中非常擅长思考，会先进行思考再给出翻译结果。你的输出格式为：
+<think>
+[思考过程]
+</think>[翻译结果]
+
+在你思考完之后，也就是</think>之后，你会给出最终的翻译即“[翻译结果]”，且[翻译结果]中不需要给出任何解释和描述，只需要提供英文的翻译结果。
+现在请你翻译以下这句英语：
+" + "The mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap."
+
+chat_response = client.chat.completions.create(
+    model=[model_name],
+    messages=[
+        {"role": "user", "content": prompt},
+    ],
+    temperature=0.1,
+    top_p=0.8,
+    max_tokens=2048,
+    extra_body={
+        "repetition_penalty": 1.05,
+    },
+)
+print("Chat response:", chat_response)
+```
+
+# ExTrans
+
+![](./images/extrans-reward-framework.png)
+
+In this work, we propose ExTrans-7B, which aims at enhancing the free translation ability of deep reasoning LLMs via **exemplar-enhanced** RL. In detail, for each training MT sample, we use DeepSeek-R1 (671B) to generate a exemplar translation, and compare the translation results of the policy model with the exemplar translations to provide rewards for the policy model.
+
+Moreover, we extend ExTrans-7B from English-to-Chinese translation into **multilingual settings** with 11 languages, *e.g.*, Chinese, English, Arabic, Czech, German, Spanish, French, Italian, Japanese, Russian and Korean. 
+
+The model checkpoints can be accessed from the following links:  
+|  | Backbone | Model Access |
+| :--: | :--: | :--: |
+| ExTrans-7B | 🤗 <a href="https://huggingface.co/Qwen/Qwen2.5-7B-Instruct">Qwen2.5-7B-Instruct</a> | 🤗 <a href="https://huggingface.co/Krystalan/ExTrans-7B">ExTrans-7B</a> |
+| mExTrans-7B | 🤗 <a href="https://huggingface.co/Qwen/Qwen2.5-7B-Instruct">Qwen2.5-7B-Instruct</a> | 🤗 <a href="https://huggingface.co/Krystalan/mExTrans-7B">mExTrans-7B</a> |
+
+
+## Inference of ExTrans
+
+deploying LLMs:
+```bash
+python3 -m vllm.entrypoints.openai.api_server --model [model_ckpt] --served-model-name [model_name]
+```
+
+calling LLMs:
+```python
+from openai import OpenAI
+# Set OpenAI's API key and API base to use vLLM's API server.
+openai_api_key = "EMPTY"
+openai_api_base = "http://localhost:8000/v1"
+
+client = OpenAI(
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+prompt = "你是一个翻译专家，擅长将英文翻译成中文。你在翻译过程中非常擅长思考，会先进行思考再给出翻译结果。你的输出格式为：
+<think>
+[思考过程]
+</think>[翻译结果]
+
+在你思考完之后，也就是</think>之后，你会给出最终的翻译即“[翻译结果]”，且[翻译结果]中不需要给出任何解释和描述，只需要提供英文的翻译结果。
+现在请你翻译以下这句英语：
+" + "The mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap."
+
+chat_response = client.chat.completions.create(
+    model=[model_name],
+    messages=[
+        {"role": "user", "content": prompt},
+    ],
+    temperature=0.1,
+    top_p=0.8,
+    max_tokens=2048,
+    extra_body={
+        "repetition_penalty": 1.05,
+    },
+)
+print("Chat response:", chat_response)
+```
+
+## Inference of mExTrans
+
+calling LLMs:
+```python
+from openai import OpenAI
+# Set OpenAI's API key and API base to use vLLM's API server.
+openai_api_key = "EMPTY"
+openai_api_base = "http://localhost:8000/v1"
+
+client = OpenAI(
+    api_key=openai_api_key,
+    base_url=openai_api_base,
+)
+
+lang2des = {
+    "ar": "阿拉伯语", # Arabic
+    "cs": "捷克语", # Czech
+    "de": "德语", # German
+    "en": "英语", # English
+    "es": "西班牙语", # Spanish
+    "fr": "法语", # French
+    "it": "意大利语", # Italian
+    "ja": "日语", # Japanese
+    "ko": "韩语", # Korean
+    "ru": "俄语", # Russian
+    "zh": "中文" # Chinese
+}
+
+current_src_lang = lang2des["en"] # set the source language
+current_trg_lang = lang2des["zh"] # set the target language
+
+current_sent = "The mother, with her feet propped up on a stool, seemed to be trying to get to the bottom of that answer, whose feminine profundity had struck her all of a heap." # the source sentence to be translated
+
+TRANS_PROMPT = "你是一个翻译专家，擅长将{current_src}翻译成{current_trg}。你在翻译过程中非常擅长思考，会先用中文进行思考再给出翻译结果。在你思考完之后，也就是</think>之后，你会给出最终的翻译，且最终的翻译结果中不需要给出任何解释和描述，只需要提供翻译结果。
+现在请你翻译以下这句{current_src}：
+{current_sent}"
 
+chat_response = client.chat.completions.create(
+    model=[model_name],
+    messages=[
+        {"role": "user", "content": TRANS_PROMPT.format(current_src=current_src_lang, current_trg=current_trg_lang, current_sent=current_sent)},
+    ],
+    temperature=0.1,
+    top_p=0.8,
+    max_tokens=2048,
+    extra_body={
+        "repetition_penalty": 1.05,
+    },
+)
+print("Chat response:", chat_response)
+```
 
+Note that, the prompt of mExTrans is slightly different from that of ExTrans.
 
+# License
+This work is licensed under cc-by-nc-sa-4.0