@lzhzzzzwill/cofos 1.1.6 → 1.1.9

package/bin/cofos.js

@@ -162,7 +162,10 @@ function main() {
   if (!existsSync(chatScript)) die(`Missing script: ${chatScript}`);
   ensureDeps(py);
 
-  console.log("  Loading COFOS 9B + RAG backend from Hugging Face … first run may take a while\n");
+  console.log("  Loading COFOS 9B + RAG backend from Hugging Face … first run may take a while");
+  if (pdfDir) console.log("  Local PDF RAG: enabled for this session");
+  else console.log("  Tip: add local papers with `cofos --pdf-dir ./new_pdfs`");
+  console.log("");
 
   const backendEnv = {
     ...process.env,
@@ -172,8 +175,8 @@ function main() {
     HF_HUB_ENABLE_HF_TRANSFER: process.env.HF_HUB_ENABLE_HF_TRANSFER || "0",
     HF_HUB_DOWNLOAD_TIMEOUT: process.env.HF_HUB_DOWNLOAD_TIMEOUT || "120",
     HF_HUB_ETAG_TIMEOUT: process.env.HF_HUB_ETAG_TIMEOUT || "30",
-    HF_HUB_VERBOSITY: process.env.HF_HUB_VERBOSITY || "warning",
-    TRANSFORMERS_VERBOSITY: process.env.TRANSFORMERS_VERBOSITY || "warning",
+    HF_HUB_VERBOSITY: process.env.HF_HUB_VERBOSITY || "error",
+    TRANSFORMERS_VERBOSITY: process.env.TRANSFORMERS_VERBOSITY || "error",
   };
   const proc = spawn(py, backendArgs, {
     stdio: ["pipe", "pipe", "pipe"],
@@ -188,8 +191,18 @@ function main() {
 
   let buf = "", ready = false, generating = false, responseOpen = false, closed = false;
   const inputQ = [];
-  const rl = createInterface({ input: process.stdin, terminal: true });
-  rl.on("line", (l) => { inputQ.push(l); });
+  const mainPrompt = C + "  cofos " + R;
+  const contPrompt = G + "  ...   " + R;
+  const rl = createInterface({
+    input: process.stdin,
+    output: process.stdout,
+    terminal: true,
+    prompt: mainPrompt,
+  });
+  rl.on("line", (l) => {
+    if (!ready || generating) inputQ.push(l);
+    else if (!handle(l)) ask();
+  });
 
   // Forward Python stderr (status / progress messages) to our stderr.
   proc.stderr.on("data", (chunk) => { process.stderr.write(chunk); });
@@ -275,9 +288,12 @@ function main() {
     ask();
   }
 
-  function ask() {
+  function ask(prompt = mainPrompt) {
     if (closed) return;
-    try { rl.question(C + "  cofos " + R, handle); } catch (_) {}
+    try {
+      rl.setPrompt(prompt);
+      rl.prompt();
+    } catch (_) {}
   }
 
   /* ─── multi-line accumulator ─── */
@@ -287,8 +303,8 @@ function main() {
     // Multi-line continuation: lines ending with "\"
     if (input.endsWith("\\")) {
       mlBuf.push(input.slice(0, -1));
-      try { rl.question(G + "  ...   " + R, handle); } catch (_) {}
-      return;
+      ask(contPrompt);
+      return true;
     }
     if (mlBuf.length > 0) {
       mlBuf.push(input);
@@ -297,20 +313,20 @@ function main() {
     }
 
     input = input.trim();
-    if (!input) { ask(); return; }
+    if (!input) return false;
 
     // Local-only commands (don't hit the backend).
-    if (/^\/(exit|quit)$/.test(input)) { proc.stdin.write("/exit\n"); proc.stdin.end(); return; }
+    if (/^\/(exit|quit)$/.test(input)) { proc.stdin.write("/exit\n"); proc.stdin.end(); return true; }
     if (/^\/(help)$/.test(input)) {
-      console.log("\n  Commands:\n    /help        Show help\n    /exit        Quit\n    /clear       Clear conversation history\n    /save        Persist conversation history\n    /info        Model and RAG info\n    /topk N      Set retrieval top-k\n    /rag on|off  Enable/disable RAG\n\n  Multi-line: end a line with \\ to continue.\n");
-      ask(); return;
+      console.log("\n  Commands:\n    /help        Show help\n    /exit        Quit\n    /clear       Clear conversation history\n    /save        Persist conversation history\n    /info        Model and RAG info\n    /topk N      Set retrieval top-k\n    /rag on|off  Enable/disable RAG\n\n  Add local papers to RAG:\n    cofos --pdf-dir ./new_pdfs\n    cofos --pdf-dir ./new_pdfs --rebuild-pdf-index\n\n  Multi-line: end a line with \\ to continue.\n");
+      return false;
     }
     if (/^\/(clear)$/.test(input)) {
       mlBuf = [];
       responseOpen = false;
       proc.stdin.write(input + "\n");
       generating = true;
-      return;
+      return true;
     }
     if (/^\/(info)$/.test(input)) {
       console.log(`\n  COFOS CLI — v${pkg.version}`);
@@ -320,7 +336,7 @@ function main() {
       console.log(`  HF:     ${backendEnv.HF_HOME}`);
       console.log("  Engine: StudentAdapterInference + transformers + PyTorch");
       console.log("  Hist:   ~/.cofos_history.jsonl\n");
-      ask(); return;
+      return false;
     }
 
     // Forward to backend. Echo a stable user turn so pasted/submitted questions remain visible in the transcript.
@@ -328,6 +344,7 @@ function main() {
     responseOpen = false;
     generating = true;
     proc.stdin.write(input + "\n");
+    return true;
   }
 }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lzhzzzzwill/cofos",
-  "version": "1.1.6",
+  "version": "1.1.9",
   "description": "COFOS 9B + RAG chat CLI with optional local PDF ingestion.",
   "type": "module",
   "bin": {

package/runtime/config/config.yaml

@@ -42,7 +42,7 @@ generation:
   temperature: 0.2
   top_p: 0.9
   max_new_tokens: 1024
-  do_sample: true
+  do_sample: false
   repetition_penalty: 1.08
   num_beams: 1
   no_repeat_ngram_size: 0
@@ -51,6 +51,7 @@ generation:
   refine_answers: false
   stream_refinement: false
   refinement_max_new_tokens: 512
+  remove_invalid_values: true
 data_processing:
   chunk_size: 768
   chunk_overlap: 50

package/runtime/config/prompt_templates.yaml

@@ -104,3 +104,16 @@ student_answer_refinement_user_prompt: 'Question: {question}
   checklist, or explanation of how you rewrote it.
 
   '
+interaction:
+  greetings:
+    hi: Hello! What would you like to know about COFs or reactive oxygen species?
+    hello: Hello! What would you like to know about COFs or reactive oxygen species?
+    hey: Hello! What would you like to know about COFs or reactive oxygen species?
+    你好: 你好！你想了解 COF、光催化还是活性氧相关问题？
+    您好: 您好！你想了解 COF、光催化还是活性氧相关问题？
+    嗨: 你好！你想了解 COF、光催化还是活性氧相关问题？
+    在吗: 在的。你想了解哪方面的 COF 或活性氧问题？
+    thanks: You're welcome.
+    thank you: You're welcome.
+    谢谢: 不客气。
+    感谢: 不客气。

package/runtime/src/inference/student_adapter_inference.py

@@ -181,6 +181,7 @@ class StudentAdapterInference:
         self.num_beams = int(generation_config.get("num_beams", 1))
         self.no_repeat_ngram_size = int(generation_config.get("no_repeat_ngram_size", 0))
         self.min_new_tokens = int(generation_config.get("min_new_tokens", 0))
+        self.remove_invalid_values = bool(generation_config.get("remove_invalid_values", True))
         self.history_max_turns = int(generation_config.get("history_max_turns", 4))
         self.refine_answers = bool(generation_config.get("refine_answers", True))
         self.stream_refinement = bool(generation_config.get("stream_refinement", True))
@@ -283,6 +284,7 @@ class StudentAdapterInference:
             "repetition_penalty": self.repetition_penalty,
             "eos_token_id": self.tokenizer.eos_token_id,
             "pad_token_id": self.tokenizer.pad_token_id,
+            "remove_invalid_values": self.remove_invalid_values,
         }
         if self.do_sample:
             kwargs["temperature"] = self.temperature
@@ -475,6 +477,7 @@ class StudentAdapterInference:
         buffer = ""
         started = False
         in_think = False
+        awaiting_initial_answer = True
 
         for chunk in chunks:
             buffer += chunk
@@ -488,6 +491,16 @@ class StudentAdapterInference:
                         break
                     buffer = buffer[end + len("</think>") :]
                     in_think = False
+                    awaiting_initial_answer = False
+                    continue
+
+                # Some tokenizers suppress the opening <think> marker but still
+                # stream the reasoning text followed by </think>.  At the start
+                # of an answer, drop everything before that closing marker too.
+                end_without_start = buffer.find("</think>")
+                if awaiting_initial_answer and end_without_start != -1:
+                    buffer = buffer[end_without_start + len("</think>") :]
+                    awaiting_initial_answer = False
                     continue
 
                 start = buffer.find("<think>")

package/scripts/chat.py

@@ -24,8 +24,8 @@ os.environ.setdefault("HF_HUB_DISABLE_XET", "1")
 os.environ.setdefault("HF_HUB_ENABLE_HF_TRANSFER", "0")
 os.environ.setdefault("HF_HUB_DOWNLOAD_TIMEOUT", "120")
 os.environ.setdefault("HF_HUB_ETAG_TIMEOUT", "30")
-os.environ.setdefault("HF_HUB_VERBOSITY", "warning")
-os.environ.setdefault("TRANSFORMERS_VERBOSITY", "warning")
+os.environ.setdefault("HF_HUB_VERBOSITY", "error")
+os.environ.setdefault("TRANSFORMERS_VERBOSITY", "error")
 
 HISTORY_FILE = os.path.join(os.path.expanduser("~"), ".cofos_history.jsonl")
 HISTORY_MAX_TURNS = 20
@@ -373,6 +373,14 @@ def ensure_runtime_data(project_root: Path, config: dict) -> None:
     print("[chat] Runtime data ready.", file=sys.stderr, flush=True)
 
 
+
+def _clean_history_content(role: str, content: str) -> str:
+    content = content.replace("COF-ROS-Reasoner", "COFOS")
+    if role == "assistant":
+        content = re.sub(r"(?is)<think>.*?</think>", "", content).strip()
+        content = re.sub(r"(?is)^.*?</think>", "", content).strip()
+    return content
+
 def _load_history() -> list:
     """Load recent conversation turns from the persistent history file.
 
@@ -398,7 +406,9 @@ def _load_history() -> list:
                     and turn.get("role") in {"user", "assistant"}
                     and isinstance(turn.get("content"), str)
                 ):
-                    turns.append(turn)
+                    cleaned = _clean_history_content(turn["role"], turn["content"])
+                    if cleaned:
+                        turns.append({"role": turn["role"], "content": cleaned})
     except OSError:
         return []
 
@@ -463,14 +473,19 @@ def main() -> None:
     import yaml
     from inference.student_adapter_inference import StudentAdapterInference, StudentRAGContext
 
-    logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
-    for noisy_logger in ("httpx", "httpcore", "huggingface_hub"):
+    debug = os.environ.get("COFOS_DEBUG", "").lower() in {"1", "true", "yes", "on"}
+    log_level = logging.INFO if debug else logging.ERROR
+    logging.basicConfig(
+        level=log_level,
+        format="%(asctime)s - %(levelname)s - %(message)s",
+        force=True,
+    )
+    for noisy_logger in ("httpx", "httpcore", "huggingface_hub", "transformers"):
         logging.getLogger(noisy_logger).setLevel(logging.WARNING)
 
     config_path = default_config_path(runtime_root) if args.config == DEFAULT_CONFIG else resolve_project_path(runtime_root, args.config)
     model_path = resolve_project_path(runtime_root, args.model_path)
 
-    debug = os.environ.get("COFOS_DEBUG", "").lower() in {"1", "true", "yes", "on"}
     if debug:
         print(f"[chat] Runtime root: {runtime_root}", file=sys.stderr, flush=True)
         print(f"[chat] Cache root: {runtime_cache_root()}", file=sys.stderr, flush=True)
@@ -571,7 +586,8 @@ def main() -> None:
             answer = "".join(answer_parts)
             gen_elapsed = time.time() - t_gen
             print("\n===DONE===", flush=True)
-            print(f"[chat] Generated in {gen_elapsed:.1f}s", file=sys.stderr, flush=True)
+            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.
@@ -580,7 +596,8 @@ def main() -> None:
             _save_turn("user", question)
             _save_turn("assistant", answer)
         except Exception as exc:
-            logging.exception("Generation failed")
+            if os.environ.get("COFOS_DEBUG", "").lower() in {"1", "true", "yes", "on"}:
+                logging.exception("Generation failed")
             print(f"\n[COFOS error] {exc}")
             print("===DONE===", flush=True)