LiquidAI/LFM3-1.7B
类型 text-generation · 框架 transformers · 下载 54,959 · 收藏 1,485
类型 text-generation · 框架 transformers · 下载 54,959 · 收藏 1,485
定位:Liquid AI 推出的 LFM3-1.7B 是面向本地 Agent 与专用应用的纯文本模型,基于 tri-hybrid 架构,适合在手机或笔记本上离线运行。 核心能力
适用人群:需要在端侧跑 Agent、工具调用或结构化输出的开发者,熟悉 transformers 与量化推理,有单卡 GPU 可做微调。 注意:官方不建议用于 Agent 编码或复杂多步推理。
位置づけ:LFM3-1.7B は Liquid AI が公開した 1.73B のテキストモデルで、tri-hybrid アーキテクチャを採用し、ローカル Agent と専用アプリ向けに設計されています。 主な機能
対象読者:ローカル環境で Agent を動かし、ツール呼び出しや構造化出力を扱う開発者。transformers 環境と単一 GPU での微調整環境が前提です。 注意:Agent コーディングや複雑な多段推論には推奨されません。
정의: LFM3-1.7B는 Liquid AI가 공개한 1.73B 텍스트 모델로, tri-hybrid 아키텍처를 기반으로 로컬 Agent와 특화 애플리케이션을 위해 설계되었습니다. 핵심 기능
대상 사용자: 로컬에서 Agent를 배포하고 도구 호출과 구조화 출력을 다루는 개발자. transformers 환경과 단일 GPU 미세 조정 조건이 필요합니다. 주의: Agent 코딩이나 복잡한 다단계 추론에는 권장되지 않습니다.
What it is: LFM3-1.7B from Liquid AI is a text model for local agents and specialized applications, built on a new tri-hybrid architecture and intended to run on phones or laptops.
Key capabilities
Who it is for: Developers deploying agents, tool calling, and structured outputs locally or on edge hardware, with a single GPU for fine-tuning and familiarity with transformers and quantized inference.
Caveats: Due to its small size, it is not recommended for agentic coding or complex multi-step reasoning.
来源 · 查看原文
libraryname: transformers license: other licensename: lfm1.0 licenselink: LICENSE pipelinetag: text-generation tags:
language:
base_model: LiquidAI/LFM3-1.7B-Base ---
LFM3-1.7B is a text model for local agents and specialized applications, based on our new tri-hybrid architecture.
Find more information about the LFM3 family in our release blog post.
The LFM3 family includes base, text, vision, audio, omni, and decision models.
| Property | Specification |
|---|---|
| Checkpoint | Post-trained, text-only language model |
| Parameters | 1.73B |
| Architecture | 24 layers: 13 gated short convolutions, 5 sliding-window attention, 6 global attention |
| Context length | 256K tokens (input + output) |
| Vocabulary | 128,000 tokens |
| Languages | English, Arabic, Chinese, French, German, Italian, Japanese, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, Polish |
| License | LFM Open License v1.0 |
LFM3-1.7B is designed for local agents, tool calling, structured outputs, and domain-specific fine-tuning. Use retrieval or search for knowledge-heavy tasks. It is not recommended for agentic coding or complex multi-step reasoning due to its small size.
Choose a checkpoint variant for your inference framework and hardware.
| Checkpoint | Framework | Description |
|---|---|---|
| BF16 | vLLM, SGLang, Transformers | BF16 weights for GPU inference. |
| FP8 | SGLang | FP8 quantization for GPU inference. |
| NVFP4 | SGLang | 4-bit floating-point quantization for GPU inference. |
| GGUF | llama.cpp, Ollama, LM Studio | Local inference on CPUs and GPUs. |
| MLX | MLX | 4-bit quantization for Apple Silicon. |
| ONNX | ONNX Runtime | Cross-platform inference with ONNX Runtime. |
Sampling parameters:
temperature=0.1,top_k=50, repetition_penalty=1.1temperature=1.0, top_p=0.95, top_k=20Transformers quick start:
from transformers import pipeline
generator = pipeline(
"text-generation",
model="LiquidAI/LFM3-1.7B",
trust_remote_code=True,
)
messages = [{"role": "user", "content": "What is C. elegans?"}]
output = generator(
messages,
do_sample=True,
temperature=0.1,
top_k=50,
repetition_penalty=1.1,
max_new_tokens=2048,
)
print(output[0]["generated_text"][-1]["content"])
For local agent setup, see the Agent Harnesses guide.
LFM3 uses this ChatML format, generated from the checkpoint's chat template:
<|startoftext|><|im_start|>system
You are a helpful assistant trained by Liquid AI.<|im_end|>
<|im_start|>user
Explain how a heat pump works.<|im_end|>
<|im_start|>assistant
<think>
Use tokenizer.apply_chat_template() with add_generation_prompt=True to format messages automatically.
Reasoning is enabled by default. You can set enable_thinking=False to disable thinking traces, but this behavior can be inconsistent, especially for reasoning-heavy tasks.
Reasoning from turns before the latest user message is removed unless preserve_thinking=True.
Pass function schemas through tools=... to apply_chat_template(). The template formats tool definitions and results automatically. Generated calls use XML:
<tool_call>
<function=get_weather>
<parameter=city>
Paris
</parameter>
</function>
</tool_call>
Execute the call in your application. Add the parsed assistant call to history as a tool_calls entry, then append the result as a tool message. Keep passing the tool schemas on subsequent requests and preserve the assistant's content and reasoning from the current tool-use turn.
function.arguments must be a Python mapping. Decode JSON strings with json.loads() before applying the template.
LFM3 supports deferred tools. Set defer_loading=True to keep tool definitions out of the prompt until your harness's search tool returns their names as tool_reference parts. The chat template then inserts their definitions into <additional_tools>.
tools = [
search_tool,
{"type": "function", "defer_loading": True, "function": get_weather_schema},
]
messages.append({
"role": "tool",
"tool_call_id": call_id,
"content": [
{"type": "text", "text": "Found 1 tool."},
{"type": "tool_reference", "name": "get_weather"},
],
})
Your harness implements and runs the search. Keep the full tools list, including deferred tools, on every request. In the system prompt, tell the model when to search:
You only have one tool right now: searchtools, which searches a catalog of many more tools. If none of your current tools can handle a request, call searchtools to find one, then call the tool it returns.
Works with vLLM, SGLang, and LFM3-compatible llama.cpp builds. For llama.cpp, use a GGUF with the current chat template or load it with --chat-template-file.
See Deferred tools and tool search for the full guide.
LFM3-1.7B is designed for easy customization for a given domain, task, or application via TRL.
LFM3-1.7B can power agentic applications that used to need 9B parameters.
| Benchmark | LFM3 1.7B |
LFM2.5 2.6B |
MiniCPM5 2B |
Gemma 4 E4B |
Qwen3.5 4B |
Qwen3.5 9B |
|---|---|---|---|---|---|---|
| Agents | ||||||
| LiveClaw | 45.60 | 14.70 | 30.37 | 20.36 | 19.94 | <u>35.77</u> |
| Claw-Eval General (EN), average | 68.64 | 62.63 | 60.74 | 56.14 | 61.96 | <u>66.53</u> |
| Claw-Eval General (EN), pass^3 | <u>51.40</u> | 37.52 | 34.58 | 22.43 | 42.99 | 51.87 |
| BrowseComp-Plus (OpenClaw subset) | <u>37.00</u> | 26.33 | 16.83 | 21.00 | 24.67 | 38.00 |
| WebArena-Verified | 38.42 | 3.02 | 9.11 | 17.06 | 21.92 | <u>28.94</u> |
| EnterpriseOps-Gym | <u>28.20</u> | 7.94 | 13.41 | 9.86 | 25.96 | 28.97 |
| AutomationBench (public) | 41.29 | 0.94 | 4.71 | 1.92 | 8.67 | <u>15.10</u> |
| Tool use | ||||||
| BFCL V4 | 71.39 | 55.89 | 62.65 | 48.25 | 55.12 | <u>62.83</u> |
| ToolSandbox | 80.36 | <u>78.04</u> | 68.16 | 69.27 | 75.40 | 76.41 |
| MCP-Atlas | 32.57 | 16.45 | 26.38 | 6.19 | <u>41.37</u> | 42.67 |
| τ³-bench (4 domains), average | <u>68.01</u> | 56.77 | 69.44 | 33.53 | 65.98 | 67.11 |
| Agentic coding | ||||||
| SWE-bench Verified (OpenCode) | <u>43.40</u> | 18.00 | 34.80 | 0.50 | 41.20 | 51.20 |
| SWE-bench Multilingual (OpenCode) | 23.00 | 8.00 | 15.67 | 0.67 | <u>23.33</u> | 35.00 |
| Terminal-Bench 2.1 (OpenCode) | 12.92 | 0.00 | 5.62 | 0.00 | <u>13.48</u> | 17.98 |
| Coding | ||||||
| LiveCodeBench v6 | 73.46 | 60.57 | 70.52 | <u>71.00</u> | 61.99 | 69.86 |
| LiveCodeBench Pro 25Q2 | <u>31.74</u> | 19.16 | 29.34 | 29.94 | 24.25 | 34.43 |
| SciCode | <u>28.47</u> | 16.32 | 27.78 | 25.00 | 23.02 | 31.60 |
| Instruction following | ||||||
| IFEval | 96.66 | <u>95.39</u> | 87.83 | 90.02 | 92.52 | 92.08 |
| IFBench | 73.03 | 60.48 | <u>62.08</u> | 44.56 | 38.16 | 48.84 |
| IFStruct | 93.72 | 82.18 | 72.65 | 78.45 | 87.05 | <u>89.20</u> |
| Multi-IF | 89.10 | <u>84.80</u> | 73.19 | 81.03 | 80.32 | 84.44 |
| Knowledge | ||||||
| GPQA Diamond | 63.37 | 57.78 | 65.15 | 57.88 | <u>75.76</u> | 80.40 |
| Humanity's Last Exam (no tools) | 7.17 | 5.88 | 8.20 | 4.77 | <u>12.84</u> | 16.41 |
| Humanity's Last Exam (with tools) | <u>13.90</u> | 9.60 | 11.50 | 6.00 | 11.60 | 17.50 |
| AA-Omniscience (public) | 0.50 | -33.17 | <u>-5.00</u> | -15.33 | -57.17 | -28.00 |
| AA-Omniscience accuracy (public) | 7.07 | 8.75 | 9.13 | 9.83 | 15.67 | <u>13.17</u> |
| AA-Omniscience non-hallucination (public) | 92.93 | 53.69 | <u>84.44</u> | 72.09 | 13.64 | 52.59 |
| Long context | ||||||
| AA-LCR | <u>52.83</u> | 6.00 | 49.34 | 39.33 | 52.67 | 62.00 |
| RULER 128K | 65.56 | 1.15 | 37.96 | 18.82 | <u>84.63</u> | 87.11 |
| Average | 49.37 | 31.55 | 39.19 | 31.74 | 39.27 | <u>47.24</u> |
<small>All models were served with vLLM in BF16 at their full context length: 262,144 tokens for LFM3-1.7B, Qwen3.5-4B, and Qwen3.5-9B, and 131,072 tokens for LFM2.5-2.6B, MiniCPM5-2B, and Gemma 4 E4B. Unless listed below, every eval uses temperature 0.6 and top_p 0.95, with no output limit other than the context length. The Average row is the unweighted mean of every score in the table.</small>
<small>
BFCL V4: greedy decoding, up to 32,768 output tokens.
ToolSandbox: greedy decoding, up to 1,024 tokens per turn.
LiveClaw: OpenClaw agent, up to 8,192 tokens per turn.
Claw-Eval: greedy decoding, 3 trials per task.
BrowseComp-Plus: OpenClaw agent, greedy decoding, up to 8,192 tokens per turn.
WebArena-Verified: up to 40 steps.
MCP-Atlas: up to 256 turns.
EnterpriseOps-Gym and AutomationBench: up to 50 turns.
SWE-bench: OpenCode agent, up to 250 turns.
Terminal-Bench 2.1: OpenCode agent, up to 100 turns.
</small>
At 32K context, LFM3-1.7B prefills at least 2.5x faster and decodes at least 1.8x faster than every compared model on all three devices, while using the least memory.

<small>All models run as Q4_0 GGUF files in llama.cpp with 32,768 tokens of context, batch size 1, 100 output tokens, and GPU offload. Prefill is the prompt processing speed and decode is the generation speed, both in tokens per second. LFM3-1.7B uses a llama.cpp build with LFM3 support, and the other models use llama.cpp release b11744.</small>
<small>
AMD Ryzen AI Max+ 395: Radeon 8060S integrated GPU, Windows, Vulkan backend.
Apple M5 Max: macOS, Metal backend.
iPhone 17 Pro: Apple A19 Pro, iOS app built on llama.cpp.
Memory: peak usage in GB (109 bytes). On AMD, it adds the host and GPU peaks from the same run, which can slightly overstate usage if the two peaks do not coincide. On Apple devices, the process footprint already includes Metal allocations.
</small>
At 1,024 concurrent requests, LFM3-1.7B can produce 3.6 billion tokens/day on one NVIDIA B300.


Pre-training: two stages build LFM3-1.7B-Base on 22T tokens in total.

Post-training: four stages develop conversation, reasoning, and tool-use capabilities.
@article{liquidAI2026LFM3,
author = {Liquid AI},
title = {Introducing LFM3: Every Modality for Specialized Intelligence},
journal = {Liquid AI Blog},
year = {2026},
note = {www.liquid.ai/blog/lfm3},
}
For questions and feedback, join Discord. For custom deployment and fine-tuning support, contact Liquid AI.