dsh-codebase-chat 0.25.3 → 0.25.6

package/dist/index.js

@@ -1620,6 +1620,10 @@ function formatHealthReport(r, lang = "fr") {
   };
   const out = [];
   out.push(`== ${t2.title} \u2014 ${basename(r.projectPath)} ==`);
+  if (r.analyzedFiles === 0) {
+    out.push(lang === "en" ? "No code files detected in this project \u2014 check the path or your .codebase-chat.json ignore rules." : "Aucun fichier de code d\xE9tect\xE9 dans ce projet \u2014 v\xE9rifie le chemin ou les r\xE8gles ignore de .codebase-chat.json.");
+    return out.join("\n");
+  }
   out.push(`${t2.score}: ${r.score}/100 (${r.grade}) \xB7 ${r.analyzedFiles} ${t2.files} \xB7 ${r.importEdges} ${t2.edges}`);
   out.push("");
   out.push(`\u25CF ${t2.cycles} (${r.cycles.length})`);
@@ -1691,6 +1695,10 @@ function formatHealthReportMd(r, lang = "fr") {
   const out = [];
   out.push(`## ${GRADE_ICON[r.grade]} ${t2.title} \u2014 \`${basename(r.projectPath)}\``);
   out.push("");
+  if (r.analyzedFiles === 0) {
+    out.push(lang === "en" ? "_No code files detected in this project \u2014 check the path or your `.codebase-chat.json` ignore rules._" : "_Aucun fichier de code d\xE9tect\xE9 dans ce projet \u2014 v\xE9rifie le chemin ou les r\xE8gles ignore de `.codebase-chat.json`._");
+    return out.join("\n");
+  }
   out.push(`**${t2.score} : ${scoreBar(r.score)} ${r.score}/100 (${r.grade})** \xB7 ${r.analyzedFiles} ${t2.files} \xB7 ${r.importEdges} ${t2.edges}`);
   out.push("");
   out.push(`### ${t2.cycles} \u2014 ${r.cycles.length}`);
@@ -2816,8 +2824,16 @@ function loadLocalModel() {
         throw new Error(`Local model not found: ${spec}`);
       }
       const { getLlama } = await importLlama();
-      const llama = await getLlama();
-      return llama.loadModel({ modelPath });
+      const cpuOnly = /^(0|off|false|cpu)$/i.test(process.env.CODEBASE_LOCAL_GPU || "");
+      const llama = await getLlama(cpuOnly ? { gpu: false } : {});
+      try {
+        return await llama.loadModel({ modelPath });
+      } catch (err) {
+        if (cpuOnly) throw err;
+        console.error(`[local-llm] GPU load failed (${err instanceof Error ? err.message : err}) \u2014 falling back to CPU. Slower, but works. Set CODEBASE_LOCAL_GPU=off to skip GPU entirely.`);
+        const cpuLlama = await getLlama({ gpu: false });
+        return cpuLlama.loadModel({ modelPath });
+      }
     })();
     modelPromise.catch(() => {
       modelPromise = null;
@@ -2834,7 +2850,13 @@ async function callLocalLlm(prompt, lang = "fr") {
       contextSequence: context.getSequence(),
       systemPrompt: lang === "en" ? "You are a senior codebase analyst. Be precise and cite files with [source: path:line]." : "Tu es un analyste codebase senior. Sois precis et cite les fichiers avec [source: chemin:ligne]."
     });
-    return await session.prompt(prompt, { temperature: 0.2, maxTokens: LOCAL_MAX_TOKENS });
+    return await session.prompt(prompt, {
+      temperature: 0.2,
+      maxTokens: LOCAL_MAX_TOKENS,
+      // Small models degenerate into loops at low temperature — penalize
+      // repeated tokens so the answer moves forward instead of echoing.
+      repeatPenalty: { lastTokens: 128, penalty: 1.2, penalizeNewLine: false }
+    });
   } finally {
     await context.dispose();
   }