chat
GitHub通过ZeroGPU调用gpt-oss-120b模型进行对话，适用于长文档、多步指令及复杂知识问答。支持系统指令注入，默认仅输出最终答案。提供liquid/deepseek/glm等替代路由以应对速度、推理或超长上下文需求。
触发场景
安装
npx skills add zerogpu/zerogpu-router --skill chat -g -y
SKILL.md
Frontmatter
{
"name": "chat",
"description": "Chat reply via ZeroGPU's default model, gpt-oss-120b (117B MoE, 131K context). Use when the user wants an answer from a ZeroGPU model rather than Claude, including longer documents, multi-step instructions, and harder general-knowledge questions. Optional system instructions via -i.",
"allowed-tools": "Bash(zerogpu chat *)",
"argument-hint": "<text> [-i <instructions>]"
}
Call the ZeroGPU chat model. $ARGUMENTS is the raw prompt. Pass it verbatim, with no escaping or quoting required (the heredoc below handles every shell metacharacter, newline, quote, and paren safely):
ZGPU_TEXT=$(cat <<'ZGPU_END_OF_INPUT'
$ARGUMENTS
ZGPU_END_OF_INPUT
)
zerogpu chat "$ZGPU_TEXT" -m gpt-oss-120b
If the user supplied system instructions, append -i "<instructions>" after the model flag.
Output is the assistant's answer as plain text. The model also produces a reasoning trace; this skill omits the CLI's -r flag so only the final answer is printed. Relay that answer as-is, without rewriting or expanding it.
For a faster, cheaper reply where quality matters less, use /zerogpu-router:chat-liquid (LFM2.5-1.2B-Instruct). For a visible reasoning trace, use /zerogpu-router:chat-thinking. For multilingual prompts, use /zerogpu-router:chat-qwen. When the input does not fit in this model's 131K context, use /zerogpu-router:chat-deepseek for code and agentic work, or /zerogpu-router:chat-glm for the largest and most capable option — both carry a 1M-token context, and glm-5.2 costs roughly 20x this skill.
Savings note: only if the command output literally contains a line starting with 💰 ZeroGPU savings, append that exact line, unchanged, as the last line of your reply. If no such line is present, say nothing about savings and do not mention or suggest /zerogpu-router:cost-savings. This note is intentionally occasional, not shown every time.
版本历史
-
cee9321
当前 2026-08-02 21:14
将默认聊天模型从LFM2.5-1.2B-Instruct更新为gpt-oss-120b；移除chat-gpt-oss技能；新增chat-deepseek和chat-glm技能以支持百万级token上下文。
- 7b9b503 2026-07-24 11:47
