first commit Qwen 0.5B

app.py

@@ -0,0 +1,302 @@
+# %%
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
+from langchain.vectorstores import Chroma
+from langchain.embeddings import HuggingFaceEmbeddings
+
+import gradio as gr
+import hanzidentifier
+import re
+
+import chinese_converter
+
+# %%
+#Load the LLM model and pipeline directly
+llm_model_name="Qwen/Qwen1.5-0.5B-Chat"
+
+#pipe = pipeline("text2text-generation", model=model)
+model = AutoModelForCausalLM.from_pretrained(
+    llm_model_name
+)
+
+tokenizer = AutoTokenizer.from_pretrained(llm_model_name)
+
+# %%
+# %%
+# loading the vector encoder
+vec_model_name = "shibing624/text2vec-base-chinese"
+
+encode_kwargs = {'normalize_embeddings': False}
+model_kwargs = {'device': 'cpu'}
+
+huggingface_embeddings= HuggingFaceEmbeddings(
+    model_name=vec_model_name,
+    model_kwargs=model_kwargs,
+    encode_kwargs = encode_kwargs
+)
+
+
+# %%
+persist_directory = 'chroma/'
+vectordb = Chroma(embedding_function=huggingface_embeddings,persist_directory=persist_directory)
+print(vectordb._collection.count())
+
+# %%
+text_input_label=["谜面","謎面","Riddle"]
+text_output_label=["谜底","謎底","Answer"]
+
+clear_label = ["清除","清除","Clear"]
+submit_label = ["提交","提交","Submit"]
+
+# %%
+# helper functions for prompt processing for this LLM
+
+def preprocess(text):
+  text = text.replace("\n", "\\n").replace("\t", "\\t")
+  return text
+
+def postprocess(text):
+  return text.replace("\\n", "\n").replace("\\t", "\t").replace('%20','  ')
+
+
+# get answer from LLM with prompt input
+def answer(input_text,context=""):
+    prompt = f"{input_text}\n提示：\n{context}\n谜底是什么？请解释。"
+    prompt = prompt.strip()
+
+    print(prompt)
+    #text = preprocess(text)
+    #out_text = pipe(text)
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": prompt}
+    ]
+    text = tokenizer.apply_chat_template(
+        messages,
+        tokenize=False,
+        add_generation_prompt=True
+    )
+    model_inputs = tokenizer([text], return_tensors="pt").to(device="cpu")
+
+    generated_ids = model.generate(
+        model_inputs.input_ids,
+        max_new_tokens=512,
+        do_sample=False,
+        top_p=0.0
+    )
+    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]
+
+    #return out_text[0]["generated_text"]
+    return response
+    #return postprocess(out_text[0]["generated_text"])
+
+# helper function for RAG
+def helper_rag(text):
+    docs_out = vectordb.similarity_search_with_relevance_scores(text,k=1)
+    #docs_out = vectordb.max_marginal_relevance_search(text,k=5,fetch_k = 20, lambda_mult = 0.5)
+    context = ""
+    for doc in docs_out:
+        if doc[1] > 0.7:
+            context += doc[0].page_content + "\n"
+
+    return context
+
+# helper function for prompt
+def helper_text(text_input,radio=None):
+    chinese_type = "simplified"
+
+    if hanzidentifier.is_traditional(text_input):
+        text_input = chinese_converter.to_simplified(text_input)
+        chinese_type = "traditional"
+
+    text_input = re.sub(r'hint',"猜",text_input,flags=re.I)
+
+    if not any(c in text_input for c in ["猜", "打"]):
+        warning = "请给一个提示，提示格式，例子：猜一水果，打一字。"
+        if chinese_type == "traditional" or radio == "繁體中文":
+            warning = chinese_converter.to_traditional(warning)
+        return warning
+                                         
+    text=f"""猜谜语：\n谜面：{text_input}
+    """
+
+    context = helper_rag(text)
+
+    output = answer(text,context=context)
+
+    print(output)
+
+    if chinese_type == "traditional":
+        output = chinese_converter.to_traditional(output)
+
+    #output = re.split(r'\s+',output)
+        
+    return output
+    
+    #return output[0]
+
+
+
+# Gradio function for configure the language of UI
+def change_language(radio,text_input,text_output,markdown, 
+                    markdown_msg1, markdown_msg2):
+    if radio == "简体中文":
+        index = 0
+        text_input_update=gr.Textbox.update(value = chinese_converter.to_simplified(text_input), label = text_input_label[index])
+        text_output_update=gr.Textbox.update(value = chinese_converter.to_simplified(text_output),label = text_output_label[index])
+        markdown_update=gr.Markdown.update(value = chinese_converter.to_simplified(markdown))
+        markdown_msg1_update=gr.Markdown.update(value = chinese_converter.to_simplified(markdown_msg1))
+        markdown_msg2_update=gr.Markdown.update(value = chinese_converter.to_simplified(markdown_msg2))
+    elif radio == "繁體中文":
+        index = 1
+        text_input_update=gr.Textbox.update(value = chinese_converter.to_traditional(text_input),label = text_input_label[index])
+        text_output_update=gr.Textbox.update(value = chinese_converter.to_traditional(text_output),label = text_output_label[index])
+        markdown_update=gr.Markdown.update(value = chinese_converter.to_traditional(markdown))
+        markdown_msg1_update=gr.Markdown.update(value = chinese_converter.to_traditional(markdown_msg1))
+        markdown_msg2_update=gr.Markdown.update(value = chinese_converter.to_traditional(markdown_msg2))
+    elif radio == "English":
+        index = 2
+        text_input_update=gr.Textbox.update(label = text_input_label[index])
+        text_output_update=gr.Textbox.update(label = text_output_label[index])
+        markdown_update=gr.Markdown.update(value = markdown)
+        markdown_msg1_update=gr.Markdown.update(value = markdown_msg1)
+        markdown_msg2_update=gr.Markdown.update(value = markdown_msg2)
+
+    else:
+        index = 0
+        text_input_update=gr.Textbox.update(label = text_input_label[index])
+        text_output_update=gr.Textbox.update(label = text_output_label[index])
+        markdown_update=gr.Markdown.update(value = chinese_converter.to_simplified(markdown))
+        markdown_msg1_update=gr.Markdown.update(value = chinese_converter.to_simplified(markdown_msg1))
+        markdown_msg2_update=gr.Markdown.update(value = chinese_converter.to_simplified(markdown_msg2))
+
+    clear_btn_update = gr.ClearButton.update(value = clear_label[index])
+    submit_btn_update = gr.Button.update(value = submit_label[index])
+
+    return [text_input_update,text_output_update,clear_btn_update,submit_btn_update,markdown_update, 
+            markdown_msg1_update ,markdown_msg2_update]
+
+
+def clear_text():
+    text_input_update=gr.Textbox.update(value=None)
+    text_output_update=gr.Textbox.update(value=None)
+
+    return [text_input_update,text_output_update]
+    
+
+# %%
+# css = """
+# #markdown { background-image: url("file/data/DSC_0105.jpg");
+#             background-size: cover;
+#           }
+# """
+
+with gr.Blocks() as demo:
+    index = 0
+    example_list = [
+        ["小家伙穿黄袍，花丛中把房造。飞到西来飞到东，人人夸他爱劳动。（猜一动物）"],
+        ["一物生来身穿三百多件衣，每天脱一件，年底剩张皮。（猜一物品）"],
+        ["A thousand threads, a million strands. Reaching the water, vanishing all at once. (Hint: natural phenomenon)"],
+        ["无底洞（猜成语）"], 
+    ]
+    radio = gr.Radio(
+        ["简体中文","繁體中文", "English"],show_label=False,value="简体中文"
+    )
+    markdown = gr.Markdown(
+            """
+            # Chinese Lantern Riddles Solver with LLM
+            ## 用大语言模型来猜灯谜 
+            """,elem_id="markdown")
+    with gr.Row():
+        with gr.Column():
+            text_input = gr.Textbox(label=text_input_label[index], 
+                         value="小家伙穿黄袍，花丛中把房造。飞到西来飞到东，人人夸他爱劳动。（猜一动物）", lines = 2)
+            with gr.Row():
+                clear_btn = gr.ClearButton(value=clear_label[index],components=[text_input])
+                submit_btn = gr.Button(value=submit_label[index], variant = "primary")
+
+            text_output = gr.Textbox(label=text_output_label[index])
+
+
+            examples = gr.Examples(
+                       examples=example_list,
+                       inputs=text_input,
+                       outputs=text_output,
+                       fn=helper_text,
+                       cache_examples=True,
+            )
+            markdown_msg1 = gr.Markdown(
+                """
+                灯谜是中华文化特色文娱活动，自北宋盛行。每年逢正月十五元宵节，将谜语贴在花灯上，让大家可一起猜谜。
+
+                Lantern riddle is a traditional Chinese cultural activity. Being popular since the Song Dynasty (960-1276), it \
+                is held in the Lantern Festival (15th day of the first lunar month). \
+                When people are viewing the flower lanterns, they can guess the riddles on the lanterns together. 
+
+
+                """
+            )
+
+        with gr.Column():
+            markdown_msg2 = gr.Markdown(
+            """
+            ![lantern](file/data/DSC_0105.jpg)
+
+            ---
+            # 声明 Disclaimer
+
+            本应用输出的文本为机器基于模型生成的结果，不代表任何人观点，请谨慎辨别和参考。请在法律允许的范围内使用。
+
+            本应用调用了 [ChatYuan-large-v2](https://huggingface.co/ClueAI/ChatYuan-large-v2) 对话语言大模型，\
+            使用本应用前请务必阅读和同意遵守其[使用授权许可证](https://huggingface.co/ClueAI/ChatYuan-large-v2/blob/main/LICENSE)。
+
+            本应用仅供非商业用途。
+
+            The outputs of this application are machine-generated with a statistical model. \
+            The outputs do not reflect any opinions of any human subjects. You must identify the outputs in caution. \
+            It is your responsbility to decide whether to accept the outputs. You must use the applicaiton in obedience to the Law.
+
+            This application utilizes [ChatYuan-large-v2](https://huggingface.co/ClueAI/ChatYuan-large-v2) \
+            Conversational Large Language Model. Before using this application, you must read and accept to follow \
+            the [LICENSE](https://huggingface.co/ClueAI/ChatYuan-large-v2/blob/main/LICENSE).
+
+            This application is for non-commercial use only.
+
+            ---
+
+            # 感谢 Acknowledgement
+
+            本应用调用了 [text2vec-base-chinese](https://huggingface.co/shibing624/text2vec-base-chinese) 生成 text vector embeddings.
+            该模型是以 [apache-2.0](https://www.apache.org/licenses/LICENSE-2.0) 发行。
+
+            This application utilizes [text2vec-base-chinese](https://huggingface.co/shibing624/text2vec-base-chinese) to generate text vector embeddings.
+            The model is released under [apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)。
+            """)
+
+            
+
+
+
+
+    submit_btn.click(fn=helper_text, inputs=[text_input,radio], outputs=text_output)
+
+    clear_btn.click(fn=clear_text,outputs=[text_input,text_output])
+    radio.change(fn=change_language,inputs=[radio,text_input,text_output,
+                                            markdown, markdown_msg1,markdown_msg2],
+                 outputs=[text_input,text_output,clear_btn,submit_btn, 
+                          markdown, markdown_msg1,markdown_msg2])
+
+    #demo = gr.Interface(fn=helper_text, inputs=text_input, outputs=text_output, 
+    #                      flagging_options=["Inappropriate"],allow_flagging="never",
+    #                      title="aaa",description="aaa",article="aaa")
+demo.queue(api_open=False)
+demo.launch(show_api=False)  
+ 
+
+# %%
+
+
+

chroma/28c81268-af8d-4e89-9744-aee894bcbde6/data_level0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0067068c3258f065668e2ad17382df3424f1c001807139881c2dc1691772ac7b
+size 28908000

chroma/28c81268-af8d-4e89-9744-aee894bcbde6/header.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b630aa796b842f9cb3af4a5f7cbc9fd3fc793047cec78a183ac3eaa83d68497
+size 100

chroma/28c81268-af8d-4e89-9744-aee894bcbde6/index_metadata.pickle

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1f937c780ce5ab483dc40c3bf8d8408b3fa3d1bf03b5d20488b0d5b8a906fde
+size 520085

chroma/28c81268-af8d-4e89-9744-aee894bcbde6/length.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a76c43e6fc007afd7f87d90b5a60a482449450491f8b307c5617b83471cf7a7
+size 36000

chroma/28c81268-af8d-4e89-9744-aee894bcbde6/link_lists.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dcde28a793504ace57c738a180d964a0c8ed59bf30194cdde6f16484f267ba06
+size 80540

chroma/c2db2536-5ceb-4ce6-b43c-7b5a8994dc6a/data_level0.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0fab3b1442c2e9319c9406cb2db197e9b806a2d4122f8c3575c62d4a0fff1dc5
+size 3212000

chroma/c2db2536-5ceb-4ce6-b43c-7b5a8994dc6a/header.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61f185c42c6b3379360fa19d3ff03ef53f86c938371ab55230c04b3aa051c549
+size 100

chroma/c2db2536-5ceb-4ce6-b43c-7b5a8994dc6a/index_metadata.pickle

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca24af71d714f0248ef6b2057c614abc943d509f5289bfdaf8d41d3f4d38d22c
+size 55974

chroma/c2db2536-5ceb-4ce6-b43c-7b5a8994dc6a/length.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9814e6d812bbb532273fec5061e2a38f55f8cab0dbe292ff0ba8d230076943ce
+size 4000

chroma/c2db2536-5ceb-4ce6-b43c-7b5a8994dc6a/link_lists.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f5dc8c4171f8df8a126112a3ebe4e91ccb0895344eff083d43ad767f6effc54
+size 8148

gradio_cached_examples/10/log.csv

@@ -0,0 +1,7 @@
+谜底,flag,username,timestamp
+这个谜面描述了一个小家伙穿着黄色的袍子，在花丛中建造房子。这个小家伙是蜜蜂，因为蜜蜂在花朵上采集花蜜，然后将这些花蜜转化为蜂蜜。所以答案是蜜蜂。,,,2024-02-19 01:00:57.388598
+谜底是日历。因为日历上的日期会随着时间的推移而改变，所以每过一年，就会有一层新的布料覆盖在上面，这就是所谓的“年轮”。因此，当最后一层布料被剥落时，就只剩下了一张没有了皮的日历。,,,2024-02-19 01:01:02.376888
+"这个谜面描述了一个自然现象，即“千条线，万条线，掉到水里看不见”。这个现象通常指的是水流在地面上形成的小水滴或小水泡，这些小水滴或小水泡会随着水流的流动而消失不见。
+
+这个谜底是雨水，因为雨水是由水滴和小水泡组成的，当它们落在地面上时，由于受到重力的作用，就会被冲走。因此，这个谜底就是雨水。",,,2024-02-19 01:01:10.248405
+这个谜面是通过“无底洞”来比喻一个无法预测或理解的、非常深邃的地方，因此可以推断出成语“深不可测”。,,,2024-02-19 01:01:13.125556