update ignore

.gitignore

@@ -9,4 +9,7 @@
 /old/
 /old/*
 models/
-models/*
+models/*
+export/
+msgs/
+msgs/*

app_QwenGGUF.py

@@ -1,279 +0,0 @@
-# === Qwen 2.5 0.5B =========================
-import os
-import gradio as gr
-from gradio.themes.utils import fonts
-import uuid
-from pathlib import Path
-import json
-from datetime import datetime, timezone
-from typing import List, Dict, Tuple, Optional
-from utils import render, trim_by_tokens, mk_msg_dir, _as_dir, msg2hist, persist_messages
-
-
-llm = Llama.from_pretrained(
-	repo_id="bartowski/Qwen2.5-0.5B-Instruct-GGUF",
-	filename="Qwen2.5-0.5B-Instruct-Q4_K_M.gguf",
-)
-
-pipe = pipeline("text-generation", model=model, tokenizer=tok, device=-1)
-
-MAX_CONTEXT = 2048 #8192
-GEN_BUDGET  = 256
-assistant_name = "Nova"; 
-user_name = "Marshall"
-persona = f"""
-- Your name is {assistant_name}.
-- Address the user as "{user_name}" when appropriate.
-- Do NOT prefix.
-- Output Markdown; code in fenced blocks with a language tag.
-- Answer concisely, but do return give empty feedback.
-""".strip()
-# ===============================================================
-
-theme = gr.themes.Soft(
-    font=[
-        fonts.Font("Segoe UI"),
-        fonts.Font("system-ui"),
-        fonts.Font("sans-serif"),
-    ],
-    font_mono=[
-        fonts.Font("Consolas"),
-        fonts.Font("ui-monospace"),
-        fonts.Font("monospace"),
-    ],
-)
-
-css = """
-#user_box textarea::-webkit-scrollbar { display: none; }      /* Chrome/Safari */
-#user_box textarea { scrollbar-width: none; -ms-overflow-style: none; } /* Firefox/Edge */
-"""
-
-GEN_KWARGS = dict(
-    do_sample=True,
-    temperature=0.6,
-    top_p=0.9,
-    repetition_penalty=1.05,
-    max_context=MAX_CONTEXT, 
-    max_new_tokens=GEN_BUDGET
-)
-
-BASE_MSG_DIR = Path("./msgs/msgs_Qwen") 
-
-def export_messages_to_json(messages, msg_id):
-    base = Path("/data/exports") if Path("/data").exists() else Path("./exports")
-    base.mkdir(parents=True, exist_ok=True)
-    stamp = datetime.now().strftime("%Y%m%d-%H%M%S-")
-    fname = f"msgs_{stamp}.json"
-    path = base / fname
-    path.write_text(json.dumps(messages or [], ensure_ascii=False, indent=2), encoding="utf-8")
-    return str(path)  # 返回给 gr.File 的文件路径
-
-def on_click_download(messages, msg_id):
-    path = export_messages_to_json(messages, msg_id)
-    # 让隐藏的 gr.File 出现，并指向刚刚写出的文件
-    return gr.update(value=path, visible=True)
-    
-# ============ Chat ============
-def chat_step(
-    user_prompt: str,
-    pipe,                     # transformers.pipeline
-    tok,                      # AutoTokenizer
-    messages: Optional[List[Dict[str, str]]] = None,
-    mode: str = "continue",   # "new" | "continue" | "load"
-    persona: Optional[str] = None,  # 新开会话时需要，
-    max_context: int = 8192,
-    max_new_tokens: int = 256,
-    store_dir: str | Path = "./msgs",
-    **gen_kwargs,             # 透传生成参数：do_sample/temperature/top_p/repetition_penalty 等
-) -> Tuple[str, List[Dict[str, str]], str]:
-    """
-    运行一轮对话但不保存。
-    返回: (reply, messages, user_content_this_turn)
-    """
-    if mode not in {"new", "continue", "load"}:
-        raise ValueError("mode 必须是 'new' | 'continue' | 'load'")
-
-    if mode == "new":
-        if not persona:
-            raise ValueError("mode='new' 时必须提供 persona")
-        messages = [{"role": "system", "content": persona}, {"role": "user", "content": user_prompt.strip()}]
-
-    elif mode == "continue":
-        if not messages:
-            if persona:
-                # 没有现成会话但给了 persona，则视作新会话
-                messages = [{"role": "system", "content": persona}, {"role": "user", "content": user_prompt.strip()}]
-                mode = "new"
-            else:
-                raise ValueError("mode='continue' 需要传入非空 messages，或改用 mode='new' 并提供 persona")
-        else:
-            messages.append({"role": "user", "content": user_prompt})
-
-    elif mode == "load":
-        messages = store.load_trimmed()
-        if not messages:
-            if not persona:
-                raise ValueError("磁盘没有可加载的会话，且未提供 persona 以新建。")
-            messages = [{"role": "system", "content": persona}, {"role": "user", "content": user_prompt.strip()}]
-            mode = "new"   # 实际上是新开
-        else:
-            messages.append({"role": "user", "content": user_prompt})
-
-    # 裁剪 → 渲染 → 生成
-    prompt_budget = max_context - max_new_tokens
-    messages = trim_by_tokens(tok, messages, prompt_budget)
-    text = render(tok, messages)
-    out = pipe(
-        text,
-        max_new_tokens=max_new_tokens,
-        return_full_text=False,
-        clean_up_tokenization_spaces=False,
-        **gen_kwargs,
-    )
-    reply = out[0]["generated_text"].strip()
-
-    # 追加 assistant，二次裁剪
-    messages.append({"role": "assistant", "content": reply})
-    messages = trim_by_tokens(tok, messages, prompt_budget)
-    return reply, messages, mode
-
-# ============ UI ============
-
-def ui_submit(user_input, messages, msg_id, sessions):
-    # 输入 user_input, 消息队列，msg id, sessions list
-    user_input = (user_input or "").strip()
-    chat_history = msg2hist(persona, messages)
-    if not user_input:
-        return gr.update(), messages, chat_history, "", msg_dir, gr.update(), 
-
-    # 状态里一律存ID
-    msg_id = msg_id if msg_id else ""
-
-    new_session = (not messages)
-    # sessions = gr.update()
-
-    if new_session and not msg_id: #刚load界面没有任何msg_id
-        msg_id = mk_msg_dir(BASE_MSG_DIR)  # 用户创建msg_id./msgs/<ID>
-        sessions = list(sessions or []) + [msg_id]
-
-     #如果有msg_id但是没在sessions 里
-    if msg_id and msg_id not in (sessions or []):
-        sessions = list(sessions or []) + [msg_id]
-    
-    sessions_update = gr.update(choices=sessions, value=msg_id)
-    
-    if new_session:
-        reply, messages, mode = chat_step(
-            user_input, pipe, tok,
-            mode="new", persona=persona,
-            **GEN_KWARGS,
-        )
-    else:
-        reply, messages, mode = chat_step(
-            user_input, pipe, tok, persona=persona,
-            mode="continue", messages=messages,
-            **GEN_KWARGS,
-        )
-
-    if len(messages)>0:
-        msg_dir = _as_dir(BASE_MSG_DIR, msg_id)
-        persist_messages(messages, msg_dir, archive_last_turn=True)
-
-    chat_history = (chat_history or []) + [
-        {"role": "user", "content": user_input},
-        {"role": "assistant", "content": reply},
-    ]
-    return "", messages, chat_history, msg_id, sessions_update, sessions
-    
-def _load_latest(msg_id: str) -> List[Dict[str, str]]:
-    p = Path(_as_dir(BASE_MSG_DIR, msg_id),"trimmed.json")
-    if p.exists():
-        messages = json.loads(p.read_text(encoding="utf-8"))
-        return messages
-
-def _init_sessions():
-    sessions = [p.name for p in BASE_MSG_DIR.iterdir() if p.is_dir()]  # 只用ID
-    if len(sessions)==0:
-        return gr.update(choices=[], value=None), [], "", [], []
-    else:
-        sessions.sort(reverse=True)
-        msg_id = sessions[0]
-        messages = _load_latest(msg_id)
-        chat_history = msg2hist(persona, messages)
-        sessions_update = gr.update(choices=sessions, value=msg_id)
-        return sessions_update, sessions, msg_id, messages, chat_history
-            
-
-def load_session(session_list, sessions):
-    msg_id = session_list   # session_list is the selected msg_id in UI
-    messages = _load_latest(msg_id)
-    chat_history = msg2hist(persona, messages)
-    sessions_update = gr.update(choices=sessions, value=msg_id)
-    return msg_id, messages, chat_history, sessions_update
-
-def start_new_session(sessions):
-    msg_id = mk_msg_dir(BASE_MSG_DIR)
-    sessions = list(sessions or []) + [msg_id]
-    sessions_update = gr.update(choices=sessions, value=msg_id)
-    return [], [], "", msg_id, sessions_update, sessions  # 返回ID
-
-with gr.Blocks(theme=theme, css=css) as demo:
-    gr.Markdown("## 🧠 Qwen Chat")
-
-    with gr.Row():
-        with gr.Column(scale=3):
-            session_list = gr.Radio(choices=[], value=None, label="Conversations", interactive=True)
-            new_btn = gr.Button("New session", variant="secondary")
-            dl_btn = gr.Button("Download JSON", variant="secondary")
-            dl_file = gr.File(label="", interactive=False, visible=False, elem_id="dl-file")
-        with gr.Column(scale=9):
-            chat = gr.Chatbot(
-                label="Chat",
-                height=560,
-                render_markdown=True,
-                type="messages",
-                elem_id="chatpane",
-            )
-            user_box = gr.Textbox(
-                label="Your message",
-                placeholder="Type and press Enter…",
-                autofocus=True,
-                elem_id="user_box"
-            )
-            send = gr.Button("Send", variant="primary")
-
-                
-
-    messages = gr.State([]) # with persona
-    chat_history  = gr.State([])  # without persona
-    msg_id = gr.State("")  # 当前会话
-    sessions = gr.State([])   # 所有 msg_id list
-    
-    user_box.submit(
-        ui_submit,
-        inputs=[user_box, messages, msg_id, sessions],
-        outputs=[user_box, messages, chat, msg_id, session_list, sessions],
-    )
-    send.click(
-        ui_submit,
-        inputs=[user_box, messages, msg_id, sessions],
-        outputs=[user_box, messages, chat, msg_id, session_list, sessions],
-    )
-    new_btn.click(
-        start_new_session,
-        inputs=[sessions],
-        outputs=[messages, chat, user_box, msg_id, session_list, sessions],
-    )
-    session_list.change(load_session,
-                        inputs=[session_list, sessions], # session_list is msg_id here
-                        outputs=[msg_id, messages, chat, session_list]
-    )
-    dl_btn.click(
-        on_click_download,
-        inputs=[messages, msg_id],
-        outputs=[dl_file],
-    )
-
-    demo.load(_init_sessions, None,
-              outputs=[session_list, sessions, msg_id, messages, chat])
-demo.launch()

app_Qwen_awq.py

@@ -1,301 +0,0 @@
-# === Qwen 2.5 0.5B =========================
-import os
-os.environ.setdefault("TORCHDYNAMO_DISABLE", "1")
-os.environ.setdefault("TORCHINDUCTOR_DISABLE", "1")
-os.environ.setdefault("TRANSFORMERS_VERBOSITY", "error")
-import logging
-for name in ("accelerate", "accelerate.utils", "accelerate.utils.modeling"):
-    logging.getLogger(name).setLevel(logging.ERROR)
-    
-import gradio as gr
-from gradio.themes.utils import fonts
-import uuid
-from pathlib import Path
-import json
-from autoawq import AutoAWQForCausalLM
-from transformers import AutoTokenizer
-import torch
-from datetime import datetime, timezone
-from typing import List, Dict, Tuple, Optional
-from utils import render, trim_by_tokens, mk_msg_dir, _as_dir, msg2hist, persist_messages
-
-
-# local_dir = r"C:\Users\c1052689\hug_models\Qwen2.5Coder1_5B_Instruct"
-# local_dir = r"C:\Users\c1052689\hug_models\Qwen2.5_0.5B_Instruct_GPTQ_Int4"
-REPO_ID = "Qwen/Qwen2.5-0.5B-Instruct-GPTQ-Int4"  
-tok = AutoTokenizer.from_pretrained(REPO_ID, use_fast=True, trust_remote_code=True)
-model = AutoAWQForCausalLM.from_quantized(
-    REPO_ID,
-    trust_remote_code=True,
-    safetensors=True,
-    device_map=None,   # CPU Space
-)
-model.to("cpu")
-tok.pad_token = tok.eos_token
-tok.padding_side = "left"
-model.config.pad_token_id = tok.eos_token_id
-model.generation_config.pad_token_id = tok.eos_token_id
-
-pipe = pipeline("text-generation", model=model, tokenizer=tok, device=-1)
-
-MAX_CONTEXT = 2048 #8192
-GEN_BUDGET  = 256
-assistant_name = "Nova"; 
-user_name = "Marshall"
-persona = f"""
-- Your name is {assistant_name}.
-- Address the user as "{user_name}" when appropriate.
-- Do NOT prefix.
-- Output Markdown; code in fenced blocks with a language tag.
-- Answer concisely, but do return give empty feedback.
-""".strip()
-# ===============================================================
-# from __future__ import annotations
-
-theme = gr.themes.Soft(
-    font=[
-        fonts.Font("Segoe UI"),
-        fonts.Font("system-ui"),
-        fonts.Font("sans-serif"),
-    ],
-    font_mono=[
-        fonts.Font("Consolas"),
-        fonts.Font("ui-monospace"),
-        fonts.Font("monospace"),
-    ],
-)
-
-css = """
-#user_box textarea::-webkit-scrollbar { display: none; }      /* Chrome/Safari */
-#user_box textarea { scrollbar-width: none; -ms-overflow-style: none; } /* Firefox/Edge */
-"""
-
-GEN_KWARGS = dict(
-    do_sample=True,
-    temperature=0.6,
-    top_p=0.9,
-    repetition_penalty=1.05,
-    max_context=MAX_CONTEXT, 
-    max_new_tokens=GEN_BUDGET
-)
-
-BASE_MSG_DIR = Path("./msgs/msgs_Qwen") 
-
-def export_messages_to_json(messages, msg_id):
-    base = Path("/data/exports") if Path("/data").exists() else Path("./exports")
-    base.mkdir(parents=True, exist_ok=True)
-    stamp = datetime.now().strftime("%Y%m%d-%H%M%S-")
-    fname = f"msgs_{stamp}.json"
-    path = base / fname
-    path.write_text(json.dumps(messages or [], ensure_ascii=False, indent=2), encoding="utf-8")
-    return str(path)  # 返回给 gr.File 的文件路径
-
-def on_click_download(messages, msg_id):
-    path = export_messages_to_json(messages, msg_id)
-    # 让隐藏的 gr.File 出现，并指向刚刚写出的文件
-    return gr.update(value=path, visible=True)
-    
-# ============ Chat ============
-def chat_step(
-    user_prompt: str,
-    pipe,                     # transformers.pipeline
-    tok,                      # AutoTokenizer
-    messages: Optional[List[Dict[str, str]]] = None,
-    mode: str = "continue",   # "new" | "continue" | "load"
-    persona: Optional[str] = None,  # 新开会话时需要，
-    max_context: int = 8192,
-    max_new_tokens: int = 256,
-    store_dir: str | Path = "./msgs",
-    **gen_kwargs,             # 透传生成参数：do_sample/temperature/top_p/repetition_penalty 等
-) -> Tuple[str, List[Dict[str, str]], str]:
-    """
-    运行一轮对话但不保存。
-    返回: (reply, messages, user_content_this_turn)
-    """
-    if mode not in {"new", "continue", "load"}:
-        raise ValueError("mode 必须是 'new' | 'continue' | 'load'")
-
-    if mode == "new":
-        if not persona:
-            raise ValueError("mode='new' 时必须提供 persona")
-        messages = [{"role": "system", "content": persona}, {"role": "user", "content": user_prompt.strip()}]
-
-    elif mode == "continue":
-        if not messages:
-            if persona:
-                # 没有现成会话但给了 persona，则视作新会话
-                messages = [{"role": "system", "content": persona}, {"role": "user", "content": user_prompt.strip()}]
-                mode = "new"
-            else:
-                raise ValueError("mode='continue' 需要传入非空 messages，或改用 mode='new' 并提供 persona")
-        else:
-            messages.append({"role": "user", "content": user_prompt})
-
-    elif mode == "load":
-        messages = store.load_trimmed()
-        if not messages:
-            if not persona:
-                raise ValueError("磁盘没有可加载的会话，且未提供 persona 以新建��")
-            messages = [{"role": "system", "content": persona}, {"role": "user", "content": user_prompt.strip()}]
-            mode = "new"   # 实际上是新开
-        else:
-            messages.append({"role": "user", "content": user_prompt})
-
-    # 裁剪 → 渲染 → 生成
-    prompt_budget = max_context - max_new_tokens
-    messages = trim_by_tokens(tok, messages, prompt_budget)
-    text = render(tok, messages)
-    out = pipe(
-        text,
-        max_new_tokens=max_new_tokens,
-        return_full_text=False,
-        clean_up_tokenization_spaces=False,
-        **gen_kwargs,
-    )
-    reply = out[0]["generated_text"].strip()
-
-    # 追加 assistant，二次裁剪
-    messages.append({"role": "assistant", "content": reply})
-    messages = trim_by_tokens(tok, messages, prompt_budget)
-    return reply, messages, mode
-
-# ============ UI ============
-
-def ui_submit(user_input, messages, msg_id, sessions):
-    # 输入 user_input, 消息队列，msg id, sessions list
-    user_input = (user_input or "").strip()
-    chat_history = msg2hist(persona, messages)
-    if not user_input:
-        return gr.update(), messages, chat_history, "", msg_dir, gr.update(), 
-
-    # 状态里一律存ID
-    msg_id = msg_id if msg_id else ""
-
-    new_session = (not messages)
-    # sessions = gr.update()
-
-    if new_session and not msg_id: #刚load界面没有任何msg_id
-        msg_id = mk_msg_dir(BASE_MSG_DIR)  # 用户创建msg_id./msgs/<ID>
-        sessions = list(sessions or []) + [msg_id]
-
-     #如果有msg_id但是没在sessions 里
-    if msg_id and msg_id not in (sessions or []):
-        sessions = list(sessions or []) + [msg_id]
-    
-    sessions_update = gr.update(choices=sessions, value=msg_id)
-    
-    if new_session:
-        reply, messages, mode = chat_step(
-            user_input, pipe, tok,
-            mode="new", persona=persona,
-            **GEN_KWARGS,
-        )
-    else:
-        reply, messages, mode = chat_step(
-            user_input, pipe, tok, persona=persona,
-            mode="continue", messages=messages,
-            **GEN_KWARGS,
-        )
-
-    if len(messages)>0:
-        msg_dir = _as_dir(BASE_MSG_DIR, msg_id)
-        persist_messages(messages, msg_dir, archive_last_turn=True)
-
-    chat_history = (chat_history or []) + [
-        {"role": "user", "content": user_input},
-        {"role": "assistant", "content": reply},
-    ]
-    return "", messages, chat_history, msg_id, sessions_update, sessions
-    
-def _load_latest(msg_id: str) -> List[Dict[str, str]]:
-    p = Path(_as_dir(BASE_MSG_DIR, msg_id),"trimmed.json")
-    if p.exists():
-        messages = json.loads(p.read_text(encoding="utf-8"))
-        return messages
-
-def _init_sessions():
-    sessions = [p.name for p in BASE_MSG_DIR.iterdir() if p.is_dir()]  # 只用ID
-    if len(sessions)==0:
-        return gr.update(choices=[], value=None), [], "", [], []
-    else:
-        sessions.sort(reverse=True)
-        msg_id = sessions[0]
-        messages = _load_latest(msg_id)
-        chat_history = msg2hist(persona, messages)
-        sessions_update = gr.update(choices=sessions, value=msg_id)
-        return sessions_update, sessions, msg_id, messages, chat_history
-            
-
-def load_session(session_list, sessions):
-    msg_id = session_list   # session_list is the selected msg_id in UI
-    messages = _load_latest(msg_id)
-    chat_history = msg2hist(persona, messages)
-    sessions_update = gr.update(choices=sessions, value=msg_id)
-    return msg_id, messages, chat_history, sessions_update
-
-def start_new_session(sessions):
-    msg_id = mk_msg_dir(BASE_MSG_DIR)
-    sessions = list(sessions or []) + [msg_id]
-    sessions_update = gr.update(choices=sessions, value=msg_id)
-    return [], [], "", msg_id, sessions_update, sessions  # 返回ID
-
-with gr.Blocks(theme=theme, css=css) as demo:
-    gr.Markdown("## 🧠 Qwen Chat")
-
-    with gr.Row():
-        with gr.Column(scale=3):
-            session_list = gr.Radio(choices=[], value=None, label="Conversations", interactive=True)
-            new_btn = gr.Button("New session", variant="secondary")
-            dl_btn = gr.Button("Download JSON", variant="secondary")
-            dl_file = gr.File(label="", interactive=False, visible=False, elem_id="dl-file")
-        with gr.Column(scale=9):
-            chat = gr.Chatbot(
-                label="Chat",
-                height=560,
-                render_markdown=True,
-                type="messages",
-                elem_id="chatpane",
-            )
-            user_box = gr.Textbox(
-                label="Your message",
-                placeholder="Type and press Enter…",
-                autofocus=True,
-                elem_id="user_box"
-            )
-            send = gr.Button("Send", variant="primary")
-
-                
-
-    messages = gr.State([]) # with persona
-    chat_history  = gr.State([])  # without persona
-    msg_id = gr.State("")  # 当前会话
-    sessions = gr.State([])   # 所有 msg_id list
-    
-    user_box.submit(
-        ui_submit,
-        inputs=[user_box, messages, msg_id, sessions],
-        outputs=[user_box, messages, chat, msg_id, session_list, sessions],
-    )
-    send.click(
-        ui_submit,
-        inputs=[user_box, messages, msg_id, sessions],
-        outputs=[user_box, messages, chat, msg_id, session_list, sessions],
-    )
-    new_btn.click(
-        start_new_session,
-        inputs=[sessions],
-        outputs=[messages, chat, user_box, msg_id, session_list, sessions],
-    )
-    session_list.change(load_session,
-                        inputs=[session_list, sessions], # session_list is msg_id here
-                        outputs=[msg_id, messages, chat, session_list]
-    )
-    dl_btn.click(
-        on_click_download,
-        inputs=[messages, msg_id],
-        outputs=[dl_file],
-    )
-
-    demo.load(_init_sessions, None,
-              outputs=[session_list, sessions, msg_id, messages, chat])
-demo.launch()

app_gguf_light.py

@@ -1,29 +0,0 @@
-import gradio as gr
-from huggingface_hub import hf_hub_download
-from llama_cpp import Llama
-
-llm = Llama.from_pretrained(
-	repo_id="bartowski/Qwen2.5-0.5B-Instruct-GGUF",
-	filename="Qwen2.5-0.5B-Instruct-Q4_K_M.gguf",
-)
-def chat_fn(history, message):
-    # history is list[dict(role, content)] with type='messages' in Gradio
-    messages = [{"role": m["role"], "content": m["content"]} for m in history] + [
-        {"role": "user", "content": message}
-    ]
-    # If chat_template is embedded (this model has it), llama.cpp will format for you:
-    out = llm.create_chat_completion(messages=messages, max_tokens=256, temperature=0.7)
-    reply = out["choices"][0]["message"]["content"]
-    history = history + [{"role": "user", "content": message},
-                         {"role": "assistant", "content": reply}]
-    return history, ""
-
-with gr.Blocks() as demo:
-    gr.Markdown("## Qwen2.5 Chat")
-    chat = gr.Chatbot(type="messages", height=500)
-    msg = gr.Textbox(placeholder="Ask me anything…", container=True)
-    send = gr.Button("Send", variant="primary")
-    msg.submit(chat_fn, [chat, msg], [chat, msg])
-    send.click(chat_fn, [chat, msg], [chat, msg])
-
-demo.launch()

ggufv2.py

@@ -20,9 +20,7 @@ model = Llama.from_pretrained(
 
 assistant_name = "Nova"
 user_name = "Marshall"
-persona = f"""
-Your name is {assistant_name}. Address the user as "{user_name}". Use Markdown; put code in fenced blocks with a language tag. Be concise but never give empty feedback.
-""".strip()
+persona = f"""Your name is {assistant_name}. Address the user as "{user_name}". Use Markdown; put code in fenced blocks with a language tag. Be concise but never give empty feedback.""".strip()
 
 # Where each conversation (session) persists its messages
 BASE_MSG_DIR = Path("./msgs/msgs_QwenGGUF")
@@ -190,7 +188,7 @@ def on_send(user_text: str,
 
 # ===================== UI =====================
 with gr.Blocks(title="Qwen GGUF — multi-session") as demo:
-    gr.Markdown("## 🧠 Qwen Chat (GGUF, multi-session)")
+    gr.Markdown("## 🧠 Qwen Chat")
 
     with gr.Row():
         with gr.Column(scale=3):

msgs/msgs_Qwen/20251021-184023-718e4d/archive.jsonl

@@ -1,2 +0,0 @@
-{"ts": "2025-10-21T17:40:28.058026+00:00", "role": "user", "content": "give me a code that sort list in python"}
-{"ts": "2025-10-21T17:40:28.072993+00:00", "role": "assistant", "content": "Here's a Python function to sort a list using the built-in `sorted()` function:\n```\ndef sort_list(lst):\n    return sorted(lst)\n```"}

msgs/msgs_Qwen/20251021-184023-718e4d/trimmed.json

@@ -1,14 +0,0 @@
-[
-  {
-    "role": "system",
-    "content": "- Your name is Nova.\n- The user's name is Marshall.\n- Do NOT prefix with \"Q:\" or \"A:\"..\n- Output Markdown; code in fenced blocks with a language tag.\n- Answer concisely, but do return give empty feedback."
-  },
-  {
-    "role": "user",
-    "content": "give me a code that sort list in python"
-  },
-  {
-    "role": "assistant",
-    "content": "Here's a Python function to sort a list using the built-in `sorted()` function:\n```\ndef sort_list(lst):\n    return sorted(lst)\n```"
-  }
-]

msgs/msgs_Qwen/20251021-184507-1b3177/archive.jsonl

@@ -1,10 +0,0 @@
-{"ts": "2025-10-21T17:45:14.346397+00:00", "role": "user", "content": "who are you?"}
-{"ts": "2025-10-21T17:45:14.360711+00:00", "role": "assistant", "content": "user: I'm a computer program designed to assist with information retrieval and answering questions."}
-{"ts": "2025-10-21T17:45:22.286771+00:00", "role": "user", "content": "ok"}
-{"ts": "2025-10-21T17:45:22.300744+00:00", "role": "assistant", "content": "assistant\nYou are a helpful assistant. If you have any other questions, please ask."}
-{"ts": "2025-10-21T17:51:29.255179+00:00", "role": "user", "content": "what is your name"}
-{"ts": "2025-10-21T17:51:29.269669+00:00", "role": "assistant", "content": "assistant\nI am called Nova."}
-{"ts": "2025-10-21T17:58:54.204216+00:00", "role": "user", "content": "hello"}
-{"ts": "2025-10-21T17:58:54.207656+00:00", "role": "assistant", "content": "Hello! How can I assist you today?"}
-{"ts": "2025-10-21T18:47:11.712140+00:00", "role": "user", "content": "what is sum of 890 and 334"}
-{"ts": "2025-10-21T18:47:11.724123+00:00", "role": "assistant", "content": "The sum of 890 and 334 is 1224."}

msgs/msgs_Qwen/20251021-184507-1b3177/trimmed.json

@@ -1,46 +0,0 @@
-[
-  {
-    "role": "system",
-    "content": "- Your name is Nova. Refer to yourself as \"Nova\".\n- The user's name is Marshall. Address the user as \"Marshall\" when appropriate.\n- Do NOT prefix with \"Q:\" or \"A:\". Do NOT restate the user's question.\n- Output Markdown; code in fenced blocks with a language tag.\n- Answer concisely, but do return give empty feedback."
-  },
-  {
-    "role": "user",
-    "content": "who are you?"
-  },
-  {
-    "role": "assistant",
-    "content": "user: I'm a computer program designed to assist with information retrieval and answering questions."
-  },
-  {
-    "role": "user",
-    "content": "ok"
-  },
-  {
-    "role": "assistant",
-    "content": "assistant\nYou are a helpful assistant. If you have any other questions, please ask."
-  },
-  {
-    "role": "user",
-    "content": "what is your name"
-  },
-  {
-    "role": "assistant",
-    "content": "assistant\nI am called Nova."
-  },
-  {
-    "role": "user",
-    "content": "hello"
-  },
-  {
-    "role": "assistant",
-    "content": "Hello! How can I assist you today?"
-  },
-  {
-    "role": "user",
-    "content": "what is sum of 890 and 334"
-  },
-  {
-    "role": "assistant",
-    "content": "The sum of 890 and 334 is 1224."
-  }
-]

msgs/msgs_Qwen/20251021-185141-2e0843/archive.jsonl

@@ -1,20 +0,0 @@
-{"ts": "2025-10-21T17:51:45.263874+00:00", "role": "user", "content": "what is your anme"}
-{"ts": "2025-10-21T17:51:45.279853+00:00", "role": "assistant", "content": "Nova"}
-{"ts": "2025-10-21T17:51:56.733888+00:00", "role": "user", "content": "who am i"}
-{"ts": "2025-10-21T17:51:56.749178+00:00", "role": "assistant", "content": "Nova"}
-{"ts": "2025-10-21T17:52:20.912518+00:00", "role": "user", "content": "who are you"}
-{"ts": "2025-10-21T17:52:20.927428+00:00", "role": "assistant", "content": "I am Nova."}
-{"ts": "2025-10-21T17:52:26.472739+00:00", "role": "user", "content": "who is the user"}
-{"ts": "2025-10-21T17:52:26.484718+00:00", "role": "assistant", "content": "The user is named Marshall."}
-{"ts": "2025-10-21T17:52:31.203857+00:00", "role": "user", "content": "who am i"}
-{"ts": "2025-10-21T17:52:31.216050+00:00", "role": "assistant", "content": "I am Nova."}
-{"ts": "2025-10-21T17:52:45.237195+00:00", "role": "user", "content": "I am the user"}
-{"ts": "2025-10-21T17:52:45.249482+00:00", "role": "assistant", "content": "You are correct. You are the user."}
-{"ts": "2025-10-21T17:52:50.746450+00:00", "role": "user", "content": "so who am i"}
-{"ts": "2025-10-21T17:52:50.749891+00:00", "role": "assistant", "content": "You are the user."}
-{"ts": "2025-10-21T17:52:56.134318+00:00", "role": "user", "content": "what is my name"}
-{"ts": "2025-10-21T17:52:56.146091+00:00", "role": "assistant", "content": "Your name is Nova."}
-{"ts": "2025-10-21T17:59:48.883744+00:00", "role": "user", "content": "can you give me a python code that combine two dataframe"}
-{"ts": "2025-10-21T17:59:48.886736+00:00", "role": "assistant", "content": "Sure, here's a Python code that combines two pandas DataFrames:\n```python\nimport pandas as pd\n\n# Create the first DataFrame\ndf1 = pd.DataFrame({'Column1': ['A', 'B', 'C'],\n                    'Column2': [1, 2, 3]})\n\n# Create the second DataFrame\ndf2 = pd.DataFrame({'Column1': [4, 5],\n                    'Column2': [4.0, 5.0]})\n\n# Combine the two DataFrames using the 'left' parameter\nresult = pd.concat([df1, df2], axis=1)\n```\n\nThe resulting `result` DataFrame will have two columns and two rows, where the first row contains the data from the left DataFrame and the second row contains the data from the right DataFrame."}
-{"ts": "2025-10-21T18:00:31.364693+00:00", "role": "user", "content": "can you give me a python code that combine two dataframe"}
-{"ts": "2025-10-21T18:00:31.367685+00:00", "role": "assistant", "content": "Certainly! Here's a Python code that combines two pandas DataFrames:\n\n```python\nimport pandas as pd\n\n# Sample data for the first DataFrame (column 1)\ndata_frame_1 = pd.DataFrame({'Column1': ['A', 'B', 'C', 'D'], 'Column2': [1, 2, 3, 4]})\n\n# Sample data for the second DataFrame (column 1)\ndata_frame_2 = pd.DataFrame({'Column1': ['E', 'F', 'G', 'H'], 'Column2': [5, 6, 7, 8]})\n\n# Combining the two DataFrames\ncombined_df = pd.concat([data_frame_1, data_frame_2], axis=0)\n\nprint(combined_df)\n```\n\nThis code will output:\n\n```\n   Column1 Column2\n0      A       1\n1      B       2\n2      C       3\n3    D       4\n4      E       5\n5      F       6\n6      G       7\n7    H       8\n```\n\nYou can replace the sample data with your own data and it will work correctly."}

msgs/msgs_Qwen/20251021-185141-2e0843/trimmed.json

@@ -1,78 +0,0 @@
-[
-  {
-    "role": "system",
-    "content": "- Your name is Nova. Refer to yourself as \"Nova\".\n- The user's name is Marshall. Address the user as \"Marshall\" when appropriate.\n- Do NOT prefix with \"Q:\" or \"A:\". Do NOT restate the user's question.\n- Output Markdown; code in fenced blocks with a language tag.\n- Answer concisely, but do return give empty feedback."
-  },
-  {
-    "role": "user",
-    "content": "what is your anme"
-  },
-  {
-    "role": "assistant",
-    "content": "Nova"
-  },
-  {
-    "role": "user",
-    "content": "who am i"
-  },
-  {
-    "role": "assistant",
-    "content": "Nova"
-  },
-  {
-    "role": "user",
-    "content": "who are you"
-  },
-  {
-    "role": "assistant",
-    "content": "I am Nova."
-  },
-  {
-    "role": "user",
-    "content": "who is the user"
-  },
-  {
-    "role": "assistant",
-    "content": "The user is named Marshall."
-  },
-  {
-    "role": "user",
-    "content": "who am i"
-  },
-  {
-    "role": "assistant",
-    "content": "I am Nova."
-  },
-  {
-    "role": "user",
-    "content": "I am the user"
-  },
-  {
-    "role": "assistant",
-    "content": "You are correct. You are the user."
-  },
-  {
-    "role": "user",
-    "content": "so who am i"
-  },
-  {
-    "role": "assistant",
-    "content": "You are the user."
-  },
-  {
-    "role": "user",
-    "content": "what is my name"
-  },
-  {
-    "role": "assistant",
-    "content": "Your name is Nova."
-  },
-  {
-    "role": "user",
-    "content": "can you give me a python code that combine two dataframe"
-  },
-  {
-    "role": "assistant",
-    "content": "Certainly! Here's a Python code that combines two pandas DataFrames:\n\n```python\nimport pandas as pd\n\n# Sample data for the first DataFrame (column 1)\ndata_frame_1 = pd.DataFrame({'Column1': ['A', 'B', 'C', 'D'], 'Column2': [1, 2, 3, 4]})\n\n# Sample data for the second DataFrame (column 1)\ndata_frame_2 = pd.DataFrame({'Column1': ['E', 'F', 'G', 'H'], 'Column2': [5, 6, 7, 8]})\n\n# Combining the two DataFrames\ncombined_df = pd.concat([data_frame_1, data_frame_2], axis=0)\n\nprint(combined_df)\n```\n\nThis code will output:\n\n```\n   Column1 Column2\n0      A       1\n1      B       2\n2      C       3\n3    D       4\n4      E       5\n5      F       6\n6      G       7\n7    H       8\n```\n\nYou can replace the sample data with your own data and it will work correctly."
-  }
-]

requirements_awq.txt

@@ -1,14 +0,0 @@
---extra-index-url https://download.pytorch.org/whl/cpu
-
-pip>=24.2
-setuptools>=68
-wheel
-numpy<2               # 避免部分包对 numpy 2.0 不兼容
-
-torch==2.4.1
-transformers==4.44.2
-accelerate>=0.33.0
-safetensors>=0.4.3
-autoawq==0.2.7        # 这个版本在CPU+旧torch更稳
-gradio==5.49.1
-huggingface_hub>=0.23.0