From add29eba08eacc6fc01cf6712eb7ca8c2cb2c5f2 Mon Sep 17 00:00:00 2001 From: binary-husky Date: Sun, 9 Feb 2025 20:26:52 +0800 Subject: [PATCH] fine tune reasoning css --- main.py | 2 +- request_llms/bridge_all.py | 2 +- request_llms/oai_std_model_template.py | 30 ++++++++++++++------------ themes/common.css | 12 +++++++++++ 4 files changed, 30 insertions(+), 16 deletions(-) diff --git a/main.py b/main.py index c250ab29..02f10b13 100644 --- a/main.py +++ b/main.py @@ -217,7 +217,7 @@ def main(): input_combo_order = ["cookies", "max_length_sl", "md_dropdown", "txt", "txt2", "top_p", "temperature", "chatbot", "history", "system_prompt", "plugin_advanced_arg"] output_combo = [cookies, chatbot, history, status] predict_args = dict(fn=ArgsGeneralWrapper(predict), inputs=[*input_combo, gr.State(True)], outputs=output_combo) - + # 提交按钮、重置按钮 multiplex_submit_btn.click( None, [multiplex_sel], None, _js="""(multiplex_sel)=>multiplex_function_begin(multiplex_sel)""") diff --git a/request_llms/bridge_all.py b/request_llms/bridge_all.py index 0679000a..cc5c1bde 100644 --- a/request_llms/bridge_all.py +++ b/request_llms/bridge_all.py @@ -1072,7 +1072,7 @@ if "zhipuai" in AVAIL_LLM_MODELS: # zhipuai 是glm-4的别名，向后兼容 }) except: logger.error(trimmed_format_exc()) -# -=-=-=-=-=-=- 幻方-深度求索大模型 -=-=-=-=-=-=- +# -=-=-=-=-=-=- 幻方-深度求索本地大模型 -=-=-=-=-=-=- if "deepseekcoder" in AVAIL_LLM_MODELS: # deepseekcoder try: from .bridge_deepseekcoder import predict_no_ui_long_connection as deepseekcoder_noui diff --git a/request_llms/oai_std_model_template.py b/request_llms/oai_std_model_template.py index 9b6a1c4c..c652a0ef 100644 --- a/request_llms/oai_std_model_template.py +++ b/request_llms/oai_std_model_template.py @@ -1,13 +1,10 @@ import json import time import traceback - import requests -from loguru import logger -# config_private.py放自己的秘密如API和代理网址 -# 读取时首先看是否存在私密的config_private配置文件（不受git管控），如果有，则覆盖原config文件 -from toolbox import get_conf, is_the_upload_folder, update_ui +from loguru import logger +from toolbox import get_conf, is_the_upload_folder, update_ui, update_ui_lastest_msg proxies, TIMEOUT_SECONDS, MAX_RETRY = get_conf( "proxies", "TIMEOUT_SECONDS", "MAX_RETRY" @@ -76,7 +73,7 @@ def decode_chunk(chunk): finish_reason = chunk["choices"][0]["finish_reason"] except: pass - return response, reasoning_content, finish_reason + return response, reasoning_content, finish_reason, str(chunk) def generate_message(input, model, key, history, max_output_token, system_prompt, temperature): @@ -162,7 +159,7 @@ def get_predict_function( 用于负责跨越线程传递已经输出的部分，大部分时候仅仅为了fancy的视觉效果，留空即可。observe_window[0]：观测窗。observe_window[1]：看门狗 """ from .bridge_all import model_info - watch_dog_patience = 5 # 看门狗的耐心，设置5秒不准咬人(咬的也不是人 + watch_dog_patience = 5 # 看门狗的耐心，设置5秒不准咬人 (咬的也不是人) if len(APIKEY) == 0: raise RuntimeError(f"APIKEY为空,请检查配置文件的{APIKEY}") if inputs == "": @@ -215,7 +212,7 @@ def get_predict_function( break except requests.exceptions.ConnectionError: chunk = next(stream_response) # 失败了，重试一次？再失败就没办法了。 - response_text, reasoning_content, finish_reason = decode_chunk(chunk) + response_text, reasoning_content, finish_reason, decoded_chunk = decode_chunk(chunk) # 返回的数据流第一次为空，继续等待 if response_text == "" and (reasoning == False or reasoning_content == "") and finish_reason != "False": continue @@ -252,9 +249,8 @@ def get_predict_function( logger.error(error_msg) raise RuntimeError("Json解析不合常规") if reasoning: - style = 'padding: 1em; line-height: 1.5; text-wrap: wrap; opacity: 0.8' paragraphs = ''.join([f'

{line}

' for line in reasoning_buffer.split('\n')]) - return f'''
{paragraphs}
\n\n''' + result + return f'''
{paragraphs}
\n\n''' + result return result def predict( @@ -348,14 +344,21 @@ def get_predict_function( gpt_reasoning_buffer = "" stream_response = response.iter_lines() + wait_counter = 0 while True: try: chunk = next(stream_response) except StopIteration: + if wait_counter != 0 and gpt_replying_buffer == "": + yield from update_ui_lastest_msg(lastmsg="模型调用失败 ...", chatbot=chatbot, history=history, msg="failed") break except requests.exceptions.ConnectionError: chunk = next(stream_response) # 失败了，重试一次？再失败就没办法了。 - response_text, reasoning_content, finish_reason = decode_chunk(chunk) + response_text, reasoning_content, finish_reason, decoded_chunk = decode_chunk(chunk) + if decoded_chunk == ': keep-alive': + wait_counter += 1 + yield from update_ui_lastest_msg(lastmsg="等待中 " + "".join(["."] * (wait_counter%10)), chatbot=chatbot, history=history, msg="waiting ...") + continue # 返回的数据流第一次为空，继续等待 if response_text == "" and (reasoning == False or reasoning_content == "") and finish_reason != "False": status_text = f"finish_reason: {finish_reason}" @@ -372,7 +375,7 @@ def get_predict_function( chunk_decoded = chunk.decode() chatbot[-1] = ( chatbot[-1][0], - f"[Local Message] {finish_reason},获得以下报错信息：\n" + f"[Local Message] {finish_reason}, 获得以下报错信息：\n" + chunk_decoded, ) yield from update_ui( @@ -390,9 +393,8 @@ def get_predict_function( if reasoning: gpt_replying_buffer += response_text gpt_reasoning_buffer += reasoning_content - style = 'padding: 1em; line-height: 1.5; text-wrap: wrap; opacity: 0.8' paragraphs = ''.join([f'

{line}

' for line in gpt_reasoning_buffer.split('\n')]) - history[-1] = f'
{paragraphs}
\n\n' + gpt_replying_buffer + history[-1] = f'
{paragraphs}
\n\n---\n\n' + gpt_replying_buffer else: gpt_replying_buffer += response_text # 如果这里抛出异常，一般是文本过长，详情见get_full_error的输出 diff --git a/themes/common.css b/themes/common.css index f2cbd32a..20bf3e8a 100644 --- a/themes/common.css +++ b/themes/common.css @@ -311,3 +311,15 @@ backdrop-filter: blur(10px); background-color: rgba(var(--block-background-fill), 0.5); } + + +.reasoning_process { + font-size: smaller; + font-style: italic; + margin: 0px; + padding: 1em; + line-height: 1.5; + text-wrap: wrap; + opacity: 0.8; +} +