@lzhzzzzwill/cofos 1.3.4 → 1.3.6

package/README.md

@@ -71,6 +71,12 @@ Disable RAG for a session:
 npx @lzhzzzzwill/cofos --no-rag
 ```
 
+Show model reasoning scaffold when emitted:
+
+```bash
+npx @lzhzzzzwill/cofos --show-reasoning
+```
+
 Ask once and exit, useful for scripts and smoke tests:
 
 ```bash
@@ -128,6 +134,7 @@ cofos \
 | `--rebuild-pdf-index` | Force PDF parsing and BM25 rebuild |
 | `--top-k` | Retrieval top-k |
 | `-p`, `--prompt` | Ask one question and exit |
+| `--show-reasoning` | Show emitted reasoning scaffold in dim text before the final answer |
 | `--max-new-tokens` | Override generation length |
 | `--no-rag` | Run without KG/BM25 retrieval |
 

package/bin/cofos.js

@@ -216,7 +216,7 @@ function createPlainUi() {
 async function main() {
   if (process.argv.includes("-h") || process.argv.includes("--help")) {
     console.log("\n  COFOS CLI — 9B + RAG chat (Willlzh/COFOS)\n");
-    console.log("  Usage:  cofos [-p QUESTION] [--model PATH] [--config PATH] [--pdf-dir DIR] [--cache-dir DIR] [--top-k N] [--no-rag]\n");
+    console.log("  Usage:  cofos [-p QUESTION] [--show-reasoning] [--model PATH] [--config PATH] [--pdf-dir DIR] [--cache-dir DIR] [--top-k N] [--no-rag]\n");
     console.log("  Commands in chat:");
     console.log("    /help        Show this help");
     console.log("    /exit        Quit");
@@ -229,6 +229,7 @@ async function main() {
     console.log("    --pdf-dir DIR          Add PDFs from DIR as extra BM25 evidence");
     console.log("    --cache-dir DIR        Store model/data cache here (default: ./.cofos)");
     console.log("    --max-new-tokens N     Override generation length");
+    console.log("    --show-reasoning       Show model reasoning scaffold in dim text when emitted");
     console.log("    -p, --prompt TEXT      Ask once and exit");
     console.log("    --rebuild-pdf-index    Force reparsing PDFs and rebuilding BM25\n");
     console.log("  Multi-line: end a line with \\ to continue on the next line.\n");
@@ -245,6 +246,7 @@ async function main() {
   let maxNewTokens = null;
   let cacheDir = process.env.COFOS_CACHE_DIR ? resolve(process.env.COFOS_CACHE_DIR) : resolve(process.cwd(), ".cofos");
   let rebuildPdfIndex = false;
+  let showReasoning = false;
   let singlePrompt = null;
   const backendArgs = [chatScript];
 
@@ -276,6 +278,8 @@ async function main() {
       cacheDir = resolve(value(++i, arg));
     } else if (arg === "--rebuild-pdf-index") {
       rebuildPdfIndex = true;
+    } else if (arg === "--show-reasoning") {
+      showReasoning = true;
     } else if (arg === "-p" || arg === "--prompt") {
       singlePrompt = value(++i, arg);
     } else if (arg === "--no-rag") {
@@ -311,6 +315,9 @@ async function main() {
     HF_HUB_ETAG_TIMEOUT: process.env.HF_HUB_ETAG_TIMEOUT || "30",
     HF_HUB_VERBOSITY: process.env.HF_HUB_VERBOSITY || "error",
     TRANSFORMERS_VERBOSITY: process.env.TRANSFORMERS_VERBOSITY || "error",
+    PYTHONUNBUFFERED: process.env.PYTHONUNBUFFERED || "1",
+    PYTHONIOENCODING: process.env.PYTHONIOENCODING || "utf-8",
+    COFOS_SHOW_REASONING: showReasoning ? "1" : (process.env.COFOS_SHOW_REASONING || "0"),
   };
   const proc = spawn(py, backendArgs, {
     stdio: ["pipe", "pipe", "pipe"],
@@ -346,6 +353,33 @@ async function main() {
     }
   }
 
+  function handleLocalOnly(rawInput) {
+    const input = String(rawInput || "").trim();
+    if (!input) return true;
+    if (input === "/") {
+      ui.print("Commands: " + SLASH_COMMANDS.map(({ cmd }) => cmd).join("  ") + "\nType a prefix such as /r or /to, then press Tab for completion.");
+      return true;
+    }
+    if (/^\/(exit|quit)$/.test(input)) { shutdown(); return true; }
+    if (/^\/(help)$/.test(input)) {
+      ui.print(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.");
+      return true;
+    }
+    if (/^\/(cls)$/.test(input)) {
+      ui.clear();
+      return true;
+    }
+    if (/^\/(info)$/.test(input)) {
+      ui.print(`\n  COFOS CLI — v${pkg.version}\n  Model:  ${modelPath} (Hugging Face)\n  RAG:    Willlzh/COFOS_data/runtime (KG + BM25) + optional --pdf-dir BM25 evidence\n  Cache:  ${cacheDir}\n  HF:     ${backendEnv.HF_HOME}\n  Engine: StudentAdapterInference + transformers + PyTorch\n  Status: ${ready ? "ready" : "loading"}\n  Hist:   ~/.cofos_history.jsonl\n`);
+      return true;
+    }
+    if (input.startsWith("/") && !/^\/(clear|save|topk)(?:\s|$)/.test(input) && !/^\/rag\s+(?:on|off)$/.test(input)) {
+      ui.print(`{red-fg}•{/red-fg} Unknown command: ${input}.\n${slashPrefixHint(input)}`);
+      return true;
+    }
+    return false;
+  }
+
   function handle(rawInput) {
     let input = String(rawInput || "");
     if (input.endsWith("\\")) {
@@ -427,8 +461,14 @@ async function main() {
   }
 
   ui.setSubmitHandler((line) => {
+    if (handleLocalOnly(line)) {
+      if (!closed) ui.focus();
+      return;
+    }
     if (!ready || generating) {
       inputQ.push(line);
+      ui.print(ready ? "• Queued until the current answer finishes." : "• Queued until the model is ready. Use /exit to quit while loading.");
+      if (!closed) ui.focus();
       return;
     }
     const started = handle(line);
@@ -476,12 +516,68 @@ async function main() {
 
   function streamPlainAssistant(text, final = false) {
     if (ui.supportsReplace) return;
+    if (!showReasoning) {
+      if (plainAssistantMode === "undecided") {
+        if (!text && !final) return;
+        plainAssistantMode = "answer";
+        ui.append(`${GR}${B}◆${R} `);
+      }
+      if (text) ui.append(text);
+      return;
+    }
+
+    plainAssistantPending += text;
     if (plainAssistantMode === "undecided") {
-      if (!text && !final) return;
-      plainAssistantMode = "answer";
-      ui.append(`${GR}${B}◆${R} `);
+      const start = plainAssistantPending.trimStart();
+      if (!start && !final) return;
+      if (start.startsWith("<think>") || reasonRe.test(start)) {
+        plainAssistantMode = "reason";
+        plainReasonBuffer = start.replace(/^<think>\s*/i, "");
+        plainReasonPrinted = 0;
+        plainAssistantPending = "";
+        ui.append(`${G}-${R} ${G}`);
+      } else {
+        plainAssistantMode = "answer";
+        plainAssistantPending = start;
+        ui.append(`${GR}${B}◆${R} `);
+      }
+    }
+
+    if (plainAssistantMode === "reason") {
+      plainReasonBuffer += plainAssistantPending;
+      plainAssistantPending = "";
+      const thinkEnd = plainReasonBuffer.indexOf("</think>");
+      const paraEnd = plainReasonBuffer.search(/\n\s*\n/);
+      const split = thinkEnd !== -1 ? thinkEnd : paraEnd;
+      if (split !== -1) {
+        const reason = plainReasonBuffer.slice(plainReasonPrinted, split).replace(/<\/think>/gi, "");
+        const restStart = thinkEnd !== -1 ? split + "</think>".length : split;
+        const rest = plainReasonBuffer.slice(restStart).trimStart();
+        if (reason) ui.append(reason);
+        ui.append(`${R}\n${GR}${B}◆${R} `);
+        plainAssistantMode = "answer";
+        plainAssistantPending = rest;
+        plainReasonBuffer = "";
+        plainReasonPrinted = 0;
+      } else {
+        const printableUntil = final ? plainReasonBuffer.length : Math.max(0, plainReasonBuffer.length - 2);
+        if (printableUntil > plainReasonPrinted) {
+          ui.append(plainReasonBuffer.slice(plainReasonPrinted, printableUntil));
+          plainReasonPrinted = printableUntil;
+        }
+        if (!final) return;
+        if (plainReasonPrinted < plainReasonBuffer.length) ui.append(plainReasonBuffer.slice(plainReasonPrinted));
+        plainReasonBuffer = "";
+        plainReasonPrinted = 0;
+        ui.append(R);
+        return;
+      }
+    }
+
+    if (plainAssistantMode === "answer" && plainAssistantPending) {
+      ui.append(plainAssistantPending);
+      plainAssistantPending = "";
     }
-    if (text) ui.append(text);
   }
 
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lzhzzzzwill/cofos",
-  "version": "1.3.4",
+  "version": "1.3.6",
   "description": "COFOS 9B + RAG chat CLI with optional local PDF ingestion.",
   "type": "module",
   "bin": {

package/runtime/src/inference/student_adapter_inference.py

@@ -489,6 +489,11 @@ class StudentAdapterInference:
     @staticmethod
     def _sanitize_stream(chunks: Any) -> Generator[str, None, None]:
         """Drop ``<think>`` blocks and leading labels from a text stream."""
+        if os.environ.get("COFOS_SHOW_REASONING", "").lower() in {"1", "true", "yes", "on"}:
+            for chunk in chunks:
+                yield chunk
+            return
+
         buffer = ""
         started = False
         in_think = False