@lzhzzzzwill/cofos 1.0.0 → 1.0.1

package/README.md

@@ -1,6 +1,6 @@
 # COFOS CLI
 
-一个本地 LLM 聊天 CLI：3D 立体欢迎界面 + Hugging Face 模型对话。
+一个本地 COFOS 模型聊天 CLI：3D 立体欢迎界面 + Hugging Face 模型对话。
 
 ```bash
 npx cofos
@@ -34,7 +34,24 @@ git clone <repo> && cd demo && npm link
 cofos
 ```
 
-首次运行会自动安装 Python 依赖并下载模型（约 30-60 秒）。
+首次运行会自动安装 Python 依赖，并从 Hugging Face 下载或读取缓存中的 `Willlzh/COFOS` 模型。9B 模型首次加载会明显慢于小模型，请耐心等待进度提示。
+
+
+## 默认模型
+
+`fwdemo` 默认加载 Hugging Face 模型：
+
+```text
+Willlzh/COFOS
+```
+
+可以在 CLI 内用 `/model <repo-or-path>` 临时切换到其它 Hugging Face 模型或本地 merged 模型目录。
+
+关联数据集：
+
+```text
+Willlzh/COFOS_data
+```
 
 ## 命令
 
@@ -93,7 +110,7 @@ cofos
 |----|------|------|
 | CLI 前端 | Node.js (ESM) | 终端 UI、子进程管理、管道通信 |
 | 模型后端 | Python + transformers | 模型加载、token 生成、流式输出 |
-| 模型 | SmolLM2-360M-Instruct (HF) | 对话推理 |
+| 模型 | Willlzh/COFOS (HF) | COFOS 专业问答推理 |
 | 通信 | stdin/stdout 管道 | 进程间消息传递 |
 | 分發 | npm | 包管理、全局命令、npx |
 
@@ -101,7 +118,7 @@ cofos
 
 - Node.js ≥ 18
 - Python 3.8+
-- 首次运行需联网（下载模型，~360MB）
+- 首次运行需联网（下载 `Willlzh/COFOS` 模型；模型体积较大）
 
 ## 开发
 

package/bin/cofos.js

@@ -92,7 +92,7 @@ function main() {
   if (!existsSync(chatScript)) { console.error("[cofos] Missing script:", chatScript); process.exit(1); }
   ensureDeps(py);
 
-  console.log("  Loading model … (30–60 s on first run)\n");
+  console.log("  Loading COFOS model … first run may take a while\n");
 
   const proc = spawn(py, [chatScript], { stdio: ["pipe", "pipe", "inherit"], encoding: "utf-8" });
 
@@ -163,7 +163,7 @@ function main() {
     if (/^\/(clear)$/.test(input)) { render(); console.log(""); ask(); return; }
     if (/^\/(info)$/.test(input)) {
       console.log("\n  COFOS CLI — v1.0.0");
-      console.log("  Model: HuggingFaceTB/SmolLM2-360M-Instruct");
+      console.log("  Model: Willlzh/COFOS");
       console.log("  Backend: transformers + PyTorch\n");
       ask(); return;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lzhzzzzwill/cofos",
-  "version": "1.0.0",
+  "version": "1.0.1",
   "description": "A local LLM chat CLI with a 3D stereoscopic welcome screen.",
   "type": "module",
   "bin": {

package/scripts/chat.py

@@ -9,21 +9,21 @@ from transformers import (
     TextIteratorStreamer,
 )
 
-MODEL_ID = "HuggingFaceTB/SmolLM2-360M-Instruct"
+MODEL_ID = "Willlzh/COFOS"
 
 
 def main() -> None:
     model_id = sys.argv[1] if len(sys.argv) > 1 else MODEL_ID
 
     print(f"[chat] Loading {model_id} …", file=sys.stderr, flush=True)
-    print(f"[chat] 30–60 s on first run (model download).", file=sys.stderr, flush=True)
+    print(f"[chat] First run may take a while while the COFOS model is downloaded or read from cache.", file=sys.stderr, flush=True)
 
-    tokenizer = AutoTokenizer.from_pretrained(model_id)
+    tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
     if tokenizer.pad_token_id is None:
         tokenizer.pad_token_id = tokenizer.eos_token_id
 
     model = AutoModelForCausalLM.from_pretrained(
-        model_id, dtype="float32", low_cpu_mem_usage=True,
+        model_id, dtype="auto", device_map="auto", trust_remote_code=True, low_cpu_mem_usage=True,
     )
     model.eval()
 
@@ -38,11 +38,11 @@ def main() -> None:
             break
         if line.startswith("/model "):
             mid = line[len("/model "):].strip()
-            tokenizer = AutoTokenizer.from_pretrained(mid)
+            tokenizer = AutoTokenizer.from_pretrained(mid, trust_remote_code=True)
             if tokenizer.pad_token_id is None:
                 tokenizer.pad_token_id = tokenizer.eos_token_id
             model = AutoModelForCausalLM.from_pretrained(
-                mid, dtype="float32", low_cpu_mem_usage=True,
+                mid, dtype="auto", device_map="auto", trust_remote_code=True, low_cpu_mem_usage=True,
             )
             model.eval()
             print("===DONE===", flush=True)