diff --git a/docs/cards/logos/recursive.png b/docs/cards/logos/recursive.png new file mode 100644 index 0000000000..ab47508938 Binary files /dev/null and b/docs/cards/logos/recursive.png differ diff --git a/docs/cookbook/autoregressive/Recursive/Flint.mdx b/docs/cookbook/autoregressive/Recursive/Flint.mdx new file mode 100644 index 0000000000..bf13ff8de6 --- /dev/null +++ b/docs/cookbook/autoregressive/Recursive/Flint.mdx @@ -0,0 +1,151 @@ +--- +title: Flint +description: "Deploy Recursive's Flint with SGLang — a 40B Mixture-of-Experts model for fast, local agentic coding with a 262K context window — on NVIDIA H200, B200, B300, GB300, and DGX Spark." +tag: NEW +--- + +## Deployment + + + + + +Flint runs on the `lmsysorg/sglang:dev-flint` Docker image: + +```bash Command +docker pull lmsysorg/sglang:dev-flint +``` + +For how to launch the image, see [Install → Method 3: Using Docker](../../../docs/get-started/install#method-3-using-docker). The **Docker** toggle in the command panel below produces the full `docker run` command; the **Python** toggle produces the inner `sglang serve ...` to run inside the container. + + + +Pick your hardware to generate the launch command. + +import { Deployment } from "/src/snippets/_deployment.jsx"; +import { config } from "/src/snippets/configs/RecursiveSI/flint.jsx"; + + + +## 1. Model Introduction + +**[Flint](https://huggingface.co/RecursiveSI/flint)** is a new open-weight **40B Mixture-of-Experts** model from Recursive, built for fast, local AI coding. Flint was developed using Recursive's Eureka Machine for automated AI research and is released under Apache 2.0. + + + + + + + + + + + + + + + + + + +
ModelTotal paramsContextUse
Flint40B (MoE)262,144Fast, local agentic coding
+ +**License:** Apache 2.0 + +**Resources:** [Hugging Face](https://huggingface.co/RecursiveSI/flint). + +## 2. Advanced Usage + +### 2.1 Reasoning + +Launch with `--reasoning-parser flint` (included in the commands above). Thinking is separated into `message.reasoning_content`, and the final answer is returned in `message.content`. + + + +```python Example +from openai import OpenAI + +client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY") + +response = client.chat.completions.create( + model="flint", + messages=[{"role": "user", "content": "What is 15% of 240?"}], +) + +message = response.choices[0].message +print("=============== Reasoning ===============") +print(message.reasoning_content) +print("=============== Answer ===============") +print(message.content) +``` + + + + + +```text Output +=============== Reasoning =============== +We need answer simple. Need maybe show calculation. User asks "What is 15% of 240?" We can answer. Need be careful maybe just 36. Let's craft concise. +=============== Answer =============== +15% of 240 is **36**. +``` + + + +### 2.2 Tool Calling + +Launch with `--tool-call-parser flint` (included in the commands above) to surface structured tool calls via `message.tool_calls`. + + + +```python Example +import json +from openai import OpenAI + +client = OpenAI(base_url="http://localhost:30000/v1", api_key="EMPTY") + +tools = [ + { + "type": "function", + "function": { + "name": "get_weather", + "description": "Get the current weather for a city", + "parameters": { + "type": "object", + "properties": {"city": {"type": "string"}}, + "required": ["city"], + }, + }, + } +] + +response = client.chat.completions.create( + model="flint", + messages=[{"role": "user", "content": "What's the weather in Paris?"}], + tools=tools, +) + +message = response.choices[0].message +print("=============== Reasoning ===============") +print(getattr(message, "reasoning_content", None)) +print("=============== Content ===============") +print(message.content) +print("=============== Tool Calls ===============") +for call in message.tool_calls or []: + print(call.function.name, json.loads(call.function.arguments)) +``` + + + + + +```text Output +=============== Reasoning =============== +The user asked for the weather in Paris, and I have a get_weather tool available that takes a city parameter. I should call it with city 'Paris' to get the current weather. +=============== Content =============== +I'll get the current weather for Paris for you. +=============== Tool Calls =============== +get_weather {'city': 'Paris'} +``` + + diff --git a/docs/cookbook/autoregressive/intro.mdx b/docs/cookbook/autoregressive/intro.mdx index 586908092c..3c5d4fa248 100644 --- a/docs/cookbook/autoregressive/intro.mdx +++ b/docs/cookbook/autoregressive/intro.mdx @@ -187,6 +187,12 @@ metatags: href="/cookbook/autoregressive/Tencent/Hy4-Preview" img="/cards/logos/tencent.png" /> +