From ed7d4943e36a4c832fda537498d3c2b6cd30f789 Mon Sep 17 00:00:00 2001 From: nik Date: Mon, 29 Jun 2026 23:17:08 +0500 Subject: [PATCH] fix: bump max_tokens to 4096, temp to 0.7 for Qwen3.6 UD --- scripts/deep_research.py | 2 +- scripts/perplexity_search.py | 2 +- scripts/research_llm_utils.py | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/scripts/deep_research.py b/scripts/deep_research.py index ec0ca11..1aee0fb 100644 --- a/scripts/deep_research.py +++ b/scripts/deep_research.py @@ -1018,7 +1018,7 @@ def synthesize_report(topic, all_articles, lang="ru"): f"Составь структурированный отчёт. Все факты подкрепляй цитатами [N]." ) - report = llm_call(system, prompt, max_tokens=4096, temperature=0.2) + report = llm_call(system, prompt, max_tokens=4096, temperature=0.7) if not report: report = "# Ошибка синтеза отчёта\n\nНе удалось сгенерировать отчёт." diff --git a/scripts/perplexity_search.py b/scripts/perplexity_search.py index d332811..9b36b61 100755 --- a/scripts/perplexity_search.py +++ b/scripts/perplexity_search.py @@ -196,7 +196,7 @@ def deduplicate(results): # ─── Sync LLM call ──────────────────────────────────────────────────────────── -def llm_chat(system, prompt, max_tokens=2000): +def llm_chat(system, prompt, max_tokens=4096): """Запрос к OpenAI-совместимому API (llama-server).""" payload = { "model": LLM_MODEL, diff --git a/scripts/research_llm_utils.py b/scripts/research_llm_utils.py index 9417023..0b92792 100644 --- a/scripts/research_llm_utils.py +++ b/scripts/research_llm_utils.py @@ -34,7 +34,7 @@ REASON_FIRST_PREFIX = ( # ─── LLM Call с retry ────────────────────────────────────────────────────────── -def llm_call_raw(system, user, max_tokens=2048, temperature=None, model=None): +def llm_call_raw(system, user, max_tokens=4096, temperature=None, model=None): """Одиночный вызов LLM. Возвращает текст или None.""" if temperature is None: temperature = LLM_TEMPERATURE