From d113218e429a9ffa1784ceea54bfa4eec5b58b2b Mon Sep 17 00:00:00 2001
From: xming521 <1223398803@qq.com>
Date: Fri, 4 Jul 2025 20:51:38 +0800
Subject: [PATCH] =?UTF-8?q?=E6=9B=B4=E6=96=B0README=E6=96=87=E6=A1=A3?=
 =?UTF-8?q?=EF=BC=9B=E4=BF=AE=E6=94=B9=E6=B8=85=E7=90=86=E7=AD=96=E7=95=A5?=
 =?UTF-8?q?=E4=B8=AD=E7=9A=84=E8=AF=84=E5=88=86=E8=A7=A3=E6=9E=90=E9=80=BB?=
 =?UTF-8?q?=E8=BE=91=E3=80=82?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 README.md                              | 10 +++++++---
 README_zh.md                           |  1 +
 weclone/core/inference/online_infer.py |  1 -
 weclone/data/clean/strategies.py       |  6 +++---
 weclone/data/models.py                 |  4 ++++
 5 files changed, 15 insertions(+), 7 deletions(-)

diff --git a/README.md b/README.md
index f80fb6c..f5f656f 100644
--- a/README.md
+++ b/README.md
@@ -55,12 +55,13 @@
 
 ### Recent Updates
 [25/06/05] Support for image modal data fine-tuning
+[25/07/10] Data source added Telegram
 
 ### Hardware Requirements
 
 The project uses Qwen2.5-VL-7B-Instruct model by default with LoRA method for SFT stage fine-tuning. You can also use other models and methods supported by [LLaMA Factory](https://github.com/hiyouga/LLaMA-Factory/tree/main#supported-models).
 
-Estimated VRAM requirements (text-only large model memory usage as follows, vision models increase based on image quantity and size): 
+Estimated VRAM requirements: 
 | Method                          | Precision |   7B  |  14B  |  30B  |   70B  |   `x`B  |
 | ------------------------------- | --------- | ----- | ----- | ----- | ------ | ------- |
 | Full (`bf16` or `fp16`)         |    32     | 120GB | 240GB | 600GB | 1200GB | `18x`GB |
@@ -162,9 +163,12 @@ weclone-cli server
 weclone-cli test-model
 ```
 
-## 🖼️ Fine-tuning Results
+## 🖼️ Results Showcase
 > [!TIP] 
-> **英文例子怎么发 More cases can be found on [XiaoHongShu](https://www.xiaohongshu.com/user/profile/628109730000000021029de4)**  
+> **We're looking for interesting examples of native English speakers chatting with WeClone! Feel free to share them with us on Twitter.
+
+？？？
+ More cases can be found on [XiaoHongShu](https://www.xiaohongshu.com/user/profile/628109730000000021029de4)**  
 
 Using the Qwen2.5VL 32B model with approximately 10,000 processed effective data samples, the loss was reduced to around 3.6:
 <details>
diff --git a/README_zh.md b/README_zh.md
index 3b2147d..5bb14e6 100644
--- a/README_zh.md
+++ b/README_zh.md
@@ -55,6 +55,7 @@
 
 ### 近期更新
 [25/06/05]支持图片模态数据微调
+[25/07/10]数据源增加Telegram
 
 ### 硬件要求
 
diff --git a/weclone/core/inference/online_infer.py b/weclone/core/inference/online_infer.py
index 9880641..d902777 100644
--- a/weclone/core/inference/online_infer.py
+++ b/weclone/core/inference/online_infer.py
@@ -29,7 +29,6 @@ class OnlineLLM:
             temperature=temperature,
             max_tokens=max_tokens,
             top_p=top_p,
-            # enable_thinking=enable_thinking   Adapt Qwen3 dynamic reasoning activation
         )
 
         return response
diff --git a/weclone/data/clean/strategies.py b/weclone/data/clean/strategies.py
index c0f1c60..06a6af5 100644
--- a/weclone/data/clean/strategies.py
+++ b/weclone/data/clean/strategies.py
@@ -10,7 +10,7 @@ from langchain_core.prompts import PromptTemplate
 from tqdm import tqdm
 
 from weclone.core.inference.online_infer import OnlineLLM
-from weclone.data.models import QaPair, QaPairScore
+from weclone.data.models import QaPair, QaPairScore, QaPairScoreWithId
 from weclone.prompts.clean_data import CLEAN_PROMPT, ONLINE_LLM_CLEAN_PROMPT
 from weclone.utils.config_models import WCMakeDatasetConfig
 from weclone.utils.log import logger
@@ -175,7 +175,7 @@ class OlineLLMCleaningStrategy(CleaningStrategy):
             # Fill template
             prompt_text = prompt_template.invoke({"qa_list": qa_list_json}).text
             try:
-                response = client.chat(prompt_text)
+                response = client.chat(prompt_text, temperature=0)
                 result_text = response.choices[0].message.content
                 # print("Model response:",result_text)
                 # If there is <think> … </think>, keep only the content after </think>
@@ -191,7 +191,7 @@ class OlineLLMCleaningStrategy(CleaningStrategy):
                     continue
 
                 for item in score_list:
-                    parsed_scores.append(QaPairScore(**item))
+                    parsed_scores.append(QaPairScoreWithId(**item))
             except Exception as e:
                 ids_in_batch = [qa["id"] for qa in qa_list]
                 logger.error(
diff --git a/weclone/data/models.py b/weclone/data/models.py
index 2361c24..f33feeb 100644
--- a/weclone/data/models.py
+++ b/weclone/data/models.py
@@ -50,6 +50,10 @@ class QaPairScore(BaseModel):
     score: int = Field(ge=1, le=5)
 
 
+class QaPairScoreWithId(QaPairScore):
+    id: int
+
+
 cut_type_data = {
     "zh_CN": [
         "Cut",