muapi Logomuapi Logo
muapi Logomuapi Logo

Llama 3.3 70B Abliterated: AI 大语言模型

通过 API 调用Llama 3.3 70B Abliterated。32K-token context, streaming and pay-per-token pricing. 可在 Muapi playground 免费试用。 Llama 3.3 70B Abliterated is an abliterated version of Meta's Llama 3.3 70B Instruct with a 32K-token context window, modified to reduce refusal behavior. Muapi exposes it through a single REST endpoint with async polling and Server-Sent Events streaming, so you can drop it into chat apps, agents and workflows without hosting GPUs. It is best suited to general chat, writing and instruction following. 按 token 计费：$1.4/M 输入 token，$1.4/M output tokens, ，无需订阅。 See also NeuralDaredevil 8B Abliterated. Browse the full lineup on the Llama Abliterated API page or the abliterated LLM API hub; related families: GLM Abliterated API and Gemma Abliterated API. 'Abliterated' and 'derestricted' describe modifications that reduce the model's refusal behavior. They do not guarantee unrestricted output, and you remain responsible for complying with applicable laws and the Muapi terms of use.

交互式模型控制

通过 API 调用Llama 3.3 70B Abliterated。32K-token context, streaming and pay-per-token pricing. 可在 Muapi playground 免费试用。

📝

概览

关于此模型

Llama 3.3 70B Abliterated is an abliterated version of Meta's Llama 3.3 70B Instruct with a 32K-token context window, modified to reduce refusal behavior. Muapi exposes it through a single REST endpoint with async polling and Server-Sent Events streaming, so you can drop it into chat apps, agents and workflows without hosting GPUs. It is best suited to general chat, writing and instruction following. 按 token 计费：$1.4/M 输入 token，$1.4/M output tokens, ，无需订阅。 See also NeuralDaredevil 8B Abliterated. Browse the full lineup on the Llama Abliterated API page or the abliterated LLM API hub; related families: GLM Abliterated API and Gemma Abliterated API. 'Abliterated' and 'derestricted' describe modifications that reduce the model's refusal behavior. They do not guarantee unrestricted output, and you remain responsible for complying with applicable laws and the Muapi terms of use.

1Chat and assistants: Build conversational apps on Llama 3.3 70B Abliterated for general chat, writing and instruction following.
2Low-refusal experimentation: Compare refusal behavior against the base model in your own evaluations, within Muapi's usage policy.
3Agents and workflows: 通过 API 调用the model from Muapi workflow nodes and the [abliterated LLM API](/abliterated-llm-api) hub examples.
💰

价格与价值

成本分析

muapiapp$1.4/M input tokens and $1.4/M output tokens

按 token 计费，按实际用量结算，请求开始时预留少量最低额度。

Fal.ai暂不可用

Llama 3.3 70B Abliterated is not offered as a hosted endpoint.

Replicate暂不可用

Llama 3.3 70B Abliterated is not offered as a hosted endpoint.

** 竞品价格根据相似模型架构和使用层级估算。

⚙️

技术详情

配置参数

提示词string

提供给模型的用户消息或指令。

默认值Summarize the trade-offs between a monolith and microservices in five bullet points.
系统提示词string

可选的系统级指令，用于引导模型行为。

默认值You are a concise, direct assistant.
最大 Token 数integer

可选的 maximum number of tokens to generate.

默认值4096
📖

实施指南

开发者文档

使用方法 Llama 3.3 70B Abliterated API

  1. 获取 API 密钥: 在 Muapi 控制台创建密钥。

  2. 发送提示词: POST prompt (plus optional system_prompt and max_tokens, up to 8192) to /api/v1/llama-3-3-70b-abliterated.

curl -X POST https://api.muapi.ai/api/v1/llama-3-3-70b-abliterated \
  -H "x-api-key: $MUAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Explain how attention works in transformers."}'
  1. 轮询或流式接收: Poll GET /api/v1/predictions/{request_id}/result until status is completed, or stream tokens live from /api/v1/llama-3-3-70b-abliterated/stream.
curl -X POST https://api.muapi.ai/api/v1/llama-3-3-70b-abliterated/stream \
  -H "x-api-key: $MUAPI_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt": "Explain how attention works in transformers."}'
  1. 查看结果: The generated text is returned in outputs. Usage is billed on actual token counts.
❓

常见问题

常见问答

什么是Llama 3.3 70B Abliterated?

Llama 3.3 70B Abliterated is an abliterated version of Meta's Llama 3.3 70B Instruct with a 32K-token context window, modified to reduce refusal behavior.

abliterated（消减拒答）是什么意思？

Abliterated 或 derestricted 模型经过修改以减少拒答行为。这会降低拒答概率，但无法完全消除拒答，也不保证输出不受限制。

如何计费：Llama 3.3 70B Abliterated

Billing is per token: $1.4/M 输入 token，$1.4/M 输出 token。 A small minimum is reserved when the request starts and reconciled to actual usage.

是否Llama 3.3 70B Abliterated支持流式输出？

支持。 Stream tokens live from `/api/v1/llama-3-3-70b-abliterated/stream` using Server-Sent Events, or poll the async endpoint.

上下文窗口有多大？

The context window is about 32K tokens. Prompt and output together must fit inside it; `max_tokens` is capped at 8192. The endpoint accepts text input only.

Is output guaranteed to be unrestricted?

不支持。 Abliterated models are less likely to refuse, but behavior varies by prompt and quality can differ from the base model. Evaluate it for your use case and follow applicable laws.