@lzhzzzzwill/cofos 1.1.9 → 1.1.13

package/bin/cofos.js

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
 import { spawn, spawnSync } from "node:child_process";
-import { createInterface } from "node:readline";
+import { createInterface, emitKeypressEvents } from "node:readline";
 import { existsSync, readFileSync } from "node:fs";
 import { dirname, join, resolve } from "node:path";
 import { fileURLToPath } from "node:url";
@@ -28,47 +28,97 @@ const S_ = [" ██████╗ ","██╔════╝ ","████
 const ART = C_.map((_, i) => [C_[i], O_[i], F_[i], O_[i], S_[i]].join("   "));
 const ART_W = ART[0].length;
 
-function shadow() {
-  let s = "  ";
-  for (const ch of ART[ART.length - 1]) s += ch !== " " ? "░" : " ";
-  return G + D + s.slice(0, -2) + "  " + R;
-}
-
 function vis(t) { return t.replace(/\x1b\[[0-9;]*m/g, "").length; }
-function fw() { return Math.max((process.stdout.columns || 100) - 2, 50); }
-
+function fw() { return Math.min(Math.max((process.stdout.columns || 110) - 2, 82), 150); }
 function top(w)  { return "╭" + "─".repeat(w) + "╮"; }
 function bot(w)  { return "╰" + "─".repeat(w) + "╯"; }
 function sep(w)  { return "├" + "─".repeat(w) + "┤"; }
-function emp(w)  { return "│" + " ".repeat(w) + "│"; }
-function cnt(t, w) {
-  const n = vis(t);
-  if (n >= w - 1) return "│" + t.slice(0, w) + "│";
-  const s = w - n, l = Math.floor(s / 2), r = s - l;
-  return "│" + " ".repeat(l) + t + " ".repeat(r) + "│";
+function vpad(t, w) { return t + " ".repeat(Math.max(0, w - vis(t))); }
+function clip(t, w) {
+  const plain = t.replace(/\x1b\[[0-9;]*m/g, "");
+  if (plain.length <= w) return t;
+  if (w <= 1) return "…";
+  return plain.slice(0, w - 1) + "…";
 }
+function cell(t, w) { return vpad(clip(t, w), w); }
+function row(left, right, lw, rw) { return "│ " + cell(left, lw) + " │ " + cell(right, rw) + " │"; }
+function full(t, w) { return "│ " + cell(t, w - 2) + " │"; }
+function empty(w) { return full("", w); }
 
 function render() {
   console.clear();
   const w = fw();
-  const l = [];
-  l.push(top(w)); l.push(emp(w));
-  if (w >= ART_W + 2) {
-    for (const r of ART) l.push(cnt(C + B + r + R, w));
-    l.push(cnt(shadow(), w));
+  const minSplit = 96;
+  const cwd = process.cwd();
+  const rag = "RAG: KG + BM25 + optional PDFs";
+  const leftTitle = C + B + "COFOS" + R;
+  const left = [
+    leftTitle,
+    G + "9B scientific RAG CLI" + R,
+    "",
+    W + "Model" + R + "  Willlzh/COFOS",
+    W + "Params" + R + "  9B",
+    W + "Path" + R + "   " + cwd,
+  ];
+  const right = [
+    W + B + "What COFOS Does" + R,
+    "Answers COF / photocatalysis / ROS questions with a local 9B model.",
+    rag,
+    "",
+    W + B + "Suggested Commands" + R,
+    "/help     show commands and PDF-RAG usage",
+    "/topk 5   change retrieval depth",
+    "/rag off  answer without retrieval",
+    "cofos --pdf-dir ./new_pdfs  add local papers",
+  ];
+
+  const lines = [top(w)];
+  if (w >= minSplit) {
+    const inner = w - 5;
+    const lw = Math.max(28, Math.floor(inner * 0.34));
+    const rw = inner - lw;
+    const n = Math.max(left.length, right.length);
+    for (let i = 0; i < n; i++) lines.push(row(left[i] || "", right[i] || "", lw, rw));
   } else {
-    l.push(cnt(C + B + "  ██████╗  ██████╗  ███████╗  ██████╗  ███████╗  " + R, w));
-    l.push(cnt(G + D + "  ░░░░░░╝  ░░░░░░╝  ░░░░░░░╝  ░░░░░░╝  ░░░░░░░╝  " + R, w));
+    for (const item of left) lines.push(full(item, w));
+    lines.push(sep(w));
+    for (const item of right) lines.push(full(item, w));
   }
-  l.push(emp(w));
-  l.push(sep(w));
-  l.push(emp(w));
-  l.push(cnt(W + B + "Welcome to COFOS CLI — 9B + RAG chat (Willlzh/COFOS)" + R, w));
-  l.push(emp(w));
-  l.push(cnt(G + "Type /help · /exit · /topk 5 · /rag off · \\ for multiline" + R, w));
-  l.push(emp(w));
-  l.push(bot(w));
-  console.log(l.join("\n"));
+  lines.push(empty(w));
+  lines.push(full(G + "Type / for commands · /exit · /topk 5 · /rag off · \\ for multiline" + R, w));
+  lines.push(bot(w));
+  console.log(lines.join("\n"));
+}
+
+/* ─── Slash commands ─── */
+const SLASH_COMMANDS = [
+  { cmd: "/help", desc: "Show help" },
+  { cmd: "/exit", desc: "Quit" },
+  { cmd: "/clear", desc: "Clear conversation history" },
+  { cmd: "/save", desc: "Save this session history" },
+  { cmd: "/info", desc: "Show model and RAG info" },
+  { cmd: "/topk 5", desc: "Set retrieval top-k" },
+  { cmd: "/rag on", desc: "Enable RAG" },
+  { cmd: "/rag off", desc: "Disable RAG" },
+];
+
+function slashMatches(line) {
+  const needle = line.trim().toLowerCase();
+  return SLASH_COMMANDS.filter(({ cmd }) => cmd.toLowerCase().startsWith(needle));
+}
+
+function slashCompletions(line) {
+  if (!line.startsWith("/")) return [[], line];
+  const hits = slashMatches(line).map(({ cmd }) => cmd);
+  return [hits.length ? hits : SLASH_COMMANDS.map(({ cmd }) => cmd), line];
+}
+
+function slashHint(line = "/") {
+  const hits = slashMatches(line);
+  const rows = (hits.length ? hits : SLASH_COMMANDS)
+    .map(({ cmd, desc }) => `    ${cmd.padEnd(10)} ${desc}`)
+    .join("\n");
+  return `\n  Commands\n${rows}\n  Press Tab to complete.\n`;
 }
 
 /* ─── helpers ─── */
@@ -198,11 +248,31 @@ function main() {
     output: process.stdout,
     terminal: true,
     prompt: mainPrompt,
+    completer: slashCompletions,
   });
+  let lastSlashHint = "";
   rl.on("line", (l) => {
+    lastSlashHint = "";
     if (!ready || generating) inputQ.push(l);
     else if (!handle(l)) ask();
   });
+  if (process.stdin.isTTY) {
+    emitKeypressEvents(process.stdin, rl);
+    process.stdin.on("keypress", () => {
+      if (!ready || generating || closed) return;
+      const line = rl.line || "";
+      if (!line.startsWith("/")) {
+        lastSlashHint = "";
+        return;
+      }
+      const key = line.trim();
+      if (key && key !== lastSlashHint) {
+        lastSlashHint = key;
+        rl.output.write(slashHint(line));
+        rl.prompt(true);
+      }
+    });
+  }
 
   // Forward Python stderr (status / progress messages) to our stderr.
   proc.stderr.on("data", (chunk) => { process.stderr.write(chunk); });
@@ -318,7 +388,7 @@ function main() {
     // Local-only commands (don't hit the backend).
     if (/^\/(exit|quit)$/.test(input)) { proc.stdin.write("/exit\n"); proc.stdin.end(); return true; }
     if (/^\/(help)$/.test(input)) {
-      console.log("\n  Commands:\n    /help        Show help\n    /exit        Quit\n    /clear       Clear conversation history\n    /save        Persist conversation history\n    /info        Model and RAG info\n    /topk N      Set retrieval top-k\n    /rag on|off  Enable/disable RAG\n\n  Add local papers to RAG:\n    cofos --pdf-dir ./new_pdfs\n    cofos --pdf-dir ./new_pdfs --rebuild-pdf-index\n\n  Multi-line: end a line with \\ to continue.\n");
+      console.log(slashHint("/") + "\n  Add local papers to RAG:\n    cofos --pdf-dir ./new_pdfs\n    cofos --pdf-dir ./new_pdfs --rebuild-pdf-index\n\n  Multi-line: end a line with \\ to continue.\n");
       return false;
     }
     if (/^\/(clear)$/.test(input)) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lzhzzzzwill/cofos",
-  "version": "1.1.9",
+  "version": "1.1.13",
   "description": "COFOS 9B + RAG chat CLI with optional local PDF ingestion.",
   "type": "module",
   "bin": {

package/runtime/src/inference/student_adapter_inference.py

@@ -2,6 +2,7 @@
 """Local student-model inference with optional confidence-gated KG/BM25 RAG."""
 
 import logging
+import os
 import re
 from pathlib import Path
 from threading import Thread
@@ -225,17 +226,31 @@ class StudentAdapterInference:
 
     @staticmethod
     def _select_dtype() -> torch.dtype:
-        """Pick the fastest dtype the current device actually supports.
+        """Pick a stable dtype for the current runtime.
 
-        bfloat16 was hardcoded, which silently falls back to a very slow
-        emulated path on CPU and on GPUs without bf16 support.
+        CUDA can safely use bf16/fp16.  On Mac/CPU, ``device_map=auto`` may place
+        layers on CPU or disk; fp16 there can produce NaN probabilities or
+        garbage tokens.  Prefer float32 outside CUDA unless the user explicitly
+        opts into another dtype with COFOS_DTYPE.
         """
+        requested = os.environ.get("COFOS_DTYPE", "auto").strip().lower()
+        aliases = {
+            "fp32": torch.float32,
+            "float32": torch.float32,
+            "fp16": torch.float16,
+            "float16": torch.float16,
+            "bf16": torch.bfloat16,
+            "bfloat16": torch.bfloat16,
+        }
+        if requested in aliases:
+            return aliases[requested]
+        if requested not in {"", "auto"}:
+            logging.warning("Ignoring unsupported COFOS_DTYPE=%s; using auto", requested)
+
         if torch.cuda.is_available():
             if torch.cuda.is_bf16_supported():
                 return torch.bfloat16
             return torch.float16
-        if torch.backends.mps.is_available():
-            return torch.float16
         return torch.float32
 
     def load_model(self) -> None:

package/scripts/chat.py

@@ -99,6 +99,11 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument("--pdf-dir", default=os.environ.get("COFOS_PDF_DIR"))
     parser.add_argument("--rebuild-pdf-index", action="store_true")
     parser.add_argument("--no-rag", action="store_true")
+    parser.add_argument(
+        "--restore-history",
+        action="store_true",
+        default=os.environ.get("COFOS_RESTORE_HISTORY", "").lower() in {"1", "true", "yes", "on"},
+    )
     return parser.parse_args()
 
 
@@ -381,6 +386,20 @@ def _clean_history_content(role: str, content: str) -> str:
         content = re.sub(r"(?is)^.*?</think>", "", content).strip()
     return content
 
+
+def _looks_like_corrupt_generation(answer: str) -> bool:
+    text = answer.strip()
+    if len(text) < 20:
+        return False
+    visible = [ch for ch in text if not ch.isspace()]
+    if not visible:
+        return True
+    alnum = sum(ch.isalnum() for ch in visible)
+    letters = sum(ch.isalpha() for ch in visible)
+    punct = sum((not ch.isalnum()) for ch in visible)
+    repeated_digits = bool(re.search(r"\d(?:\s*\d){8,}", text))
+    return (letters / len(visible) < 0.20 and punct / len(visible) > 0.35) or repeated_digits
+
 def _load_history() -> list:
     """Load recent conversation turns from the persistent history file.
 
@@ -520,8 +539,8 @@ def main() -> None:
     print("===READY===", flush=True)
 
     top_k = args.top_k
-    history = _load_history()
-    if history:
+    history = _load_history() if args.restore_history else []
+    if history and os.environ.get("COFOS_DEBUG", "").lower() in {"1", "true", "yes", "on"}:
         print(f"[chat] Restored {len(history) // 2} previous turns from history.",
               file=sys.stderr, flush=True)
 
@@ -589,12 +608,14 @@ def main() -> None:
             if os.environ.get("COFOS_DEBUG", "").lower() in {"1", "true", "yes", "on"}:
                 print(f"[chat] Generated in {gen_elapsed:.1f}s", file=sys.stderr, flush=True)
 
-            # Persist both turns only once the answer completed, so an
-            # interrupted generation cannot leave a dangling user turn behind.
-            history.append({"role": "user", "content": question})
-            history.append({"role": "assistant", "content": answer})
-            _save_turn("user", question)
-            _save_turn("assistant", answer)
+            # Keep history in-memory for the current session. Persist only when
+            # the user explicitly runs /save, so a bad generation cannot poison
+            # future launches.
+            if _looks_like_corrupt_generation(answer):
+                print("[COFOS warning] The generated text looked corrupted and was not kept in history.", file=sys.stderr, flush=True)
+            else:
+                history.append({"role": "user", "content": question})
+                history.append({"role": "assistant", "content": answer})
         except Exception as exc:
             if os.environ.get("COFOS_DEBUG", "").lower() in {"1", "true", "yes", "on"}:
                 logging.exception("Generation failed")