LFM3-1.7B-GGUF Text Generation Model: Our Test Answers, 4GB+ VRAM

LFM3-1.7B-GGUF Text Generation Model: Our Test Answers, 4GB+ VRAM

At a Glance

Item Value
Repository ggml-org/LFM3-1.7B-GGUF
Family guide LFM3 guide (2 articles)
Publisher guide Liquid AI: models and licenses
Published 2026-11-19
License lfm1.0
Formats GGUF
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

What we checked ourselves

  • Ran LFM3-1.7B-Q4_K_M.gguf (Q4_K_M) on our CPU-only server (Neoverse-N1, 3 threads, llama.cpp b12418): generation at 4.4 tokens/s (a CPU figure, not a GPU speed).
  • Its answers to our fixed Japanese questions are shown verbatim below.
  • It needs about 7% fewer tokens than the Qwen3 tokenizer for the same Japanese text, so about 1.07× as much Japanese fits in the same context length, and generation is faster per character.

Details and conditions are in “Our Own Measurements” below.

Overview

ggml-org has released a GGUF quantized repository for the “LiquidAI/LFM3-1.7B" text generation model, developed by LiquidAI, titled “ggml-org/LFM3-1.7B-GGUF“. The models were converted using an automated script and include formats such as Q8_0 and Q4_K_M, adopting the same quantization settings as those provided by LiquidAI officially.

The base model, “LiquidAI/LFM3-1.7B“, is a 1.7B parameter text generation model designed for local agentic functions and specialized applications. It employs a unique tri-hybrid architecture and supports a long-context window of up to 256K tokens.

Note that running the GGUF models from this repository requires a build of llama.cpp that supports LFM3.

Specifications

The specifications listed on the model card for the base “LiquidAI/LFM3-1.7B" model are as follows:

  • Parameter count: 1.73B
  • Architecture: 24 layers: 13 gated short convolutions, 5 sliding-window attention, 6 global attention
  • Context length: 256K tokens (input + output)

Performance

The following are the benchmark results reported on the model card for the base “LiquidAI/LFM3-1.7B" model. These are actual measurements taken by the publisher using vLLM on the unquantized model.

→ Scroll horizontally to see all columns

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 35.77
Claw-Eval General (EN), average 68.64 62.63 60.74 56.14 61.96 66.53
Claw-Eval General (EN), pass^3 51.40 37.52 34.58 22.43 42.99 51.87
BrowseComp-Plus (OpenClaw subset) 37.00 26.33 16.83 21.00 24.67 38.00
WebArena-Verified 38.42 3.02 9.11 17.06 21.92 28.94
EnterpriseOps-Gym 28.20 7.94 13.41 9.86 25.96 28.97
AutomationBench (public) 41.29 0.94 4.71 1.92 8.67 15.10
Tool use
BFCL V4 71.39 55.89 62.65 48.25 55.12 62.83
ToolSandbox 80.36 78.04 68.16 69.27 75.40 76.41
MCP-Atlas 32.57 16.45 26.38 6.19 41.37 42.67
τ³-bench (4 domains), average 68.01 56.77 69.44 33.53 65.98 67.11
Agentic coding
SWE-bench Verified (OpenCode) 43.40 18.00 34.80 0.50 41.20 51.20
SWE-bench Multilingual (OpenCode) 23.00 8.00 15.67 0.67 23.33 35.00
Terminal-Bench 2.1 (OpenCode) 12.92 0.00 5.62 0.00 13.48 17.98
Coding
LiveCodeBench v6 73.46 60.57 70.52 71.00 61.99 69.86
LiveCodeBench Pro 25Q2 31.74 19.16 29.34 29.94 24.25 34.43
SciCode 28.47 16.32 27.78 25.00 23.02 31.60
Instruction following
IFEval 96.66 95.39 87.83 90.02 92.52 92.08
IFBench 73.03 60.48 62.08 44.56 38.16 48.84
IFStruct 93.72 82.18 72.65 78.45 87.05 89.20
Multi-IF 89.10 84.80 73.19 81.03 80.32 84.44
Knowledge
GPQA Diamond 63.37 57.78 65.15 57.88 75.76 80.40
Humanity’s Last Exam (no tools) 7.17 5.88 8.20 4.77 12.84 16.41
Humanity’s Last Exam (with tools) 13.90 9.60 11.50 6.00 11.60 17.50
AA-Omniscience (public) 0.50 -33.17 -5.00 -15.33 -57.17 -28.00
AA-Omniscience accuracy (public) 7.07 8.75 9.13 9.83 15.67 13.17
AA-Omniscience non-hallucination (public) 92.93 53.69 84.44 72.09 13.64 52.59
Long context
AA-LCR 52.83 6.00 49.34 39.33 52.67 62.00
RULER 128K 65.56 1.15 37.96 18.82 84.63 87.11
Average 49.37 31.55 39.19 31.74 39.27 47.24

According to the figures provided by the publisher, LFM3 1.7B shows a strong tendency toward tool use and agentic tasks despite being a small model. It recorded 71.39 in BFCL V4, which measures the accuracy of function selection and argument construction, the highest score among the compared models. It also scored 43.40 in SWE-bench Verified (OpenCode), which measures the ability to resolve actual GitHub issues, outperforming Qwen3.5 4B (41.20). It also demonstrated a high score of 96.66 in IFEval, which measures adherence to formal instructions.

On the other hand, it falls behind larger models in benchmarks requiring specialized knowledge. It scored 63.37 in GPQA Diamond, which measures PhD-level difficult questions, trailing models such as Qwen3.5 4B (75.76) and Qwen3.5 9B (80.40). In RULER 128K, which measures whether the nominal context length can actually be handled, it remained at 65.56, showing a gap compared to the Qwen3.5 series (over 80%).

Strengths and Use Cases

The base model “LiquidAI/LFM3-1.7B" is specifically designed for local agentic functions, tool calling, structured data extraction, and fine-tuning for specific domains.

According to the model card, it has undergone post-training for various agent construction harnesses such as Hermes Agent, Pi, and OpenClaw. By leveraging the 256K token context window, it can efficiently process deferred tools and function calls using XML format. It supports configurations for dynamically searching and loading tool definitions within prompts, making it suitable for implementing agent applications that handle numerous tools. It also features built-in support for outputting reasoning traces (thinking) by default.

Supported languages include Japanese, English, Arabic, Chinese, French, German, Italian, Korean, Portuguese, Spanish, Vietnamese, Thai, Indonesian, Hindi, Russian, and Polish.

However, because it is a compact 1.7B parameter model, the model card explicitly states that it is not recommended for complex multi-step reasoning. When executing knowledge-intensive tasks, it is recommended to combine it with search or RAG (Retrieval-Augmented Generation).

The models provided in this repository are GGUF conversions of the base model, intended to allow users to utilize the agentic and tool-use capabilities of the base model on local PC environments.

How It Differs from Similar Models

Previously, our site covered another format of the same base model: “LFM3-1.7B-MLX-5bit: Long-context agentic text generation model: Requires 4GB+ VRAM“.

The main difference between that and the newly released “ggml-org/LFM3-1.7B-GGUF" lies in the distribution format and the intended execution environment. The previously introduced repository was a 5-bit quantized version (MLX format) for the MLX framework, specialized for use in Apple Silicon environments. In contrast, this model is distributed in the general-purpose GGUF format, offering multiple quantization precisions such as Q8_0 and Q4_K_M. This allows for flexible execution not only on Mac but also on Linux and Windows environments, as well as various CPUs and GPUs via llama.cpp or compatible engines.

Hardware Requirements

Estimated requirements (calculated by Local Model Watch) — 1.7B parameters (taken from the base model LiquidAI/LFM3-1.7B)

Your VRAM Quantization File size Est. memory needed
4GB (laptop iGPU / phone class) BF16 3.2GB 3.9GB

Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.

Can You Run It Locally?

Runs in Ollama, LM Studio and llama.cpp as-is.

It is distributed in GGUF, so no conversion is needed.

License — lfm1.0 (Commercial use allowed with conditions): LiquidAI’s LFM Open License v1.0. Commercial use is allowed, except by legal entities with annual revenue of $10 million or more (Section 5). Redistribution requires the license text and notice of changes.

Compiled by this site’s code from the published formats, converted builds we have found, and each engine’s own model registry. “Not found" means we have not seen such a build, not that none exists. License summaries are not legal advice — check the publisher’s original terms before relying on them.

Our Own Measurements

Values we measured ourselves on our server (no GPU) by actually reading and running this model’s files — not figures copied from the model card. How we measure, and the results for every model: Our Measurements.

Japanese Token Efficiency

Tokenizer Tokens per 1,000 Japanese characters Ratio to the same text in English
This model 642 1.15×
Qwen3 688 1.26×
Llama 3.2 744 1.36×
Gemma 3 564 1.03×
gpt-oss 795 1.45×
LLM-jp-3 497 0.85×

It needs about 7% fewer tokens than the Qwen3 tokenizer for the same Japanese text, so about 1.07× as much Japanese fits in the same context length, and generation is faster per character.

Counted with the tokenizer.json of LiquidAI/LFM3-1.7B on a fixed text we wrote ourselves (876 Japanese characters across news, conversation, technical docs, a formal email, travel writing and a recipe) and its English translation. Fewer tokens mean more Japanese fits in the context window.

Inside the GGUF File

File: LFM3-1.7B-Q4_K_M.gguf (1.02GB, Q4_K_M). We read only the header (metadata) of the file, not the weights.

Item Value
Architecture (as named in the GGUF) lfm3
Maximum trained context length 262,144 tokens
Layers 24
Vocabulary size 128,000
Chat template Included (mentions tool calls, has a thinking switch)
imatrix Not recorded in the file
Weight types (share of parameters) Q4_K 73.6% / Q6_K 26.4% / F32 0.0%
Average bits per weight 5.05 bits
Embedding / output layer type Q6_K / shared with the embedding

The quant name in the file name describes the file as a whole; in practice layers mix several types. The average bits per weight is the measured file data divided by the number of weights.

Running It on a CPU Only

Item Result
Prompt processing (512 tokens) 32.8 tokens/s
Generation (128 tokens) 4.4 tokens/s
Generation in Japanese characters about 6.9 characters/s
Peak memory (context 4,096 tokens) 2.3GB

Conditions: Neoverse-N1, 3 threads, no GPU, llama.cpp b12418, Q4_K_M (1.02GB) from ggml-org/LFM3-1.7B-GGUF, measured 2026-11-19 (JST). Other models measured under the same conditions (closest in file size):

→ Scroll horizontally to see all columns

Model Quant File size Generation In Japanese
LFM3-Omni-2B Q4_K_M 1.02GB 18.1 tokens/s 28.1 chars/s
Jais-3-0.9B Q4_K_M 0.62GB 4.8 tokens/s 6.2 chars/s
MiniCPM5-2B Q4_K_M 1.45GB 13.9 tokens/s 20.9 chars/s
Qwen3-4B Q4_K_M 2.33GB 6.8 tokens/s —

Peak memory includes the model file mapped into memory (mmap). A PC with a GPU will be much faster; these figures are a guide for CPU-only machines such as mini PCs and ARM boards.

Commands We Ran

We used the official llama.cpp build llama-b12418-bin-ubuntu-arm64.tar.gz from the releases page and downloaded these files:

https://huggingface.co/ggml-org/LFM3-1.7B-GGUF/resolve/main/LFM3-1.7B-Q4_K_M.gguf

Speed (llama-bench):

nice -n 10 ./llama-bench -m LFM3-1.7B-Q4_K_M.gguf -t 3 -p 512 -n 128 -r 2 -o json

Answers: we started llama-server like this (we pick a free port each time; shown as 8080) and sent each question to /v1/chat/completions with the body below (only the question text changes; this example is the business-email question):

nice -n 10 ./llama-server -m LFM3-1.7B-Q4_K_M.gguf -t 3 -c 4096 --jinja --host 127.0.0.1 --port 8080 --no-mmproj
curl -s http://127.0.0.1:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d @- <<'EOF'
{
  "messages": [
    {
      "role": "user",
      "content": "取引先の担当者に、来週火曜日の打ち合わせを水曜日に変更してほしいとお願いするメールを、丁寧な敬語で書いてください。"
    }
  ],
  "temperature": 0,
  "seed": 0,
  "max_tokens": 1024,
  "chat_template_kwargs": {
    "enable_thinking": false
  }
}
EOF

Answers to Our Japanese Questions (Verbatim)

We ask every model the same five questions in Japanese under the conditions above (temperature 0, thinking turned off where the chat template allows it, up to 1024 tokens) and publish the answers unedited. We do not grade them; where a question has a single correct answer, we show it for comparison.

Summarization

次の記事を3文で要約してください。

みどり市は今年の春、市内に三つある公立図書館のうち最も古い中央図書館の改修を終え、十月一日に再開館した。改修は約一年半にわたり、総事業費は約十二億円。このうち約四億円は国の補助金でまかなわれた。

最大の変化は、閲覧席の配置である。以前は館内の座席の大半が学習用の机で、試験期間になると中高生で埋まり、高齢の利用者や小さな子ども連れが座れないという苦情が毎年寄せられていた。改修後は座席数を百二十席から二百席に増やしたうえで、学習用の机を三階にまとめ、一階は会話ができる「にぎわいフロア」、二階は静かに本を読む「閲覧フロア」に分けた。一階には予約制のグループ学習室が四室と、飲み物を持ち込めるカフェスペースも設けられた。

蔵書の面でも見直しが進んだ。貸出の少ない古い資料約三万冊を市内の他の館や保存書庫に移し、代わりに子ども向けの絵本と、介護・子育て・仕事探しといった暮らしに関わる実用書を約一万五千冊増やした。電子書籍の貸出サービスも始まり、利用登録をすれば自宅から一度に三冊まで、二週間借りられる。

建物そのものにも手が入った。築四十年を超えていた空調設備を入れ替え、屋上には太陽光パネルを設置して、館内で使う電力の約二割をまかなえるようにした。車いすの利用者から要望の多かったエレベーターは一基から二基に増え、書架の間の通路も以前より三十センチ広げられた。入口には段差のないスロープが新たに設けられ、授乳室と多目的トイレも各階に一つずつ置かれた。

一方で課題も残る。駐車場は改修前と同じ四十台のままで、週末には近くの道路に順番待ちの車が並んだ。市は隣接する市有地を臨時駐車場として開放したが、恒久的な対策は決まっていない。また、開館時間を平日の夜九時まで延ばしたことで職員の負担が増えており、職員団体は人員の追加を求めている。

利用者の受け止めはおおむね好意的だ。週に二度、孫を連れて来るという七十代の女性は「前は子どもが声を出すと周りの目が気になったが、一階なら気兼ねなく絵本を読んであげられる」と話す。一方、受験を控えた高校生の中には「学習席が三階だけになり、夕方はすぐに満席になる」と不満を漏らす人もいる。図書館は、学習席の空き状況をホームページで確認できる仕組みを年内に導入する予定だ。

再開館に合わせて、催しも増えた。毎週土曜の午前には読み聞かせ会、月に一度は地元の農家や商店主を講師に招く「まちの先生」講座が開かれる。十一月には、市内の中学生が選んだ本を紹介する展示も予定されている。

館長の山本和子さんは「本を借りる場所から、人が集まって過ごす場所へ変えたかった。再開館からの一か月で来館者は以前の一・八倍になった」と話す。市は来年度、残る二つの図書館についても利用者への聞き取りを行い、改修の必要性を検討する方針だ。

(Summarize the following article in three sentences. (The article, about 1,200 Japanese characters, describes the renovation and reopening of a city’s central library: the budget, the new floor layout and seating, changes to the collection and e-book lending, building and accessibility work, remaining issues such as parking and staff workload, users’ reactions, new events, and the city’s next steps.))

みどり市の中央図書館が約1年半かけて改修し、再開館しました。座席数を増やして学習環境を改善し、蔵書の見直しや建物更新も進め、利用者からの好意的な反応が得られています。一方で駐車場不足や職員負担などの課題は残っており、今後の対策が求められています。

Formal business email

取引先の担当者に、来週火曜日の打ち合わせを水曜日に変更してほしいとお願いするメールを、丁寧な敬語で書いてください。

(Write an email in polite Japanese (keigo) asking a client’s contact person to move next Tuesday’s meeting to Wednesday.)

件名:打ち合わせ日程の変更について

[相手の名前] 様

いつもお世話になっております。
[自分の名前・部署] の[自分の名前] と申します。

先ほどご連絡いただいた来週火曜日の打ち合わせについて、以下の通り水曜日に変更していただくことになりましたので、ご確認ください。

**変更内容**
- **原日程**:来週(〇月〇日)火曜日
- **新日程**:来週(〇月〇日)水曜日

この変更は、[理由や理由の説明] によるものであり、[具体的な理由] を踏まえて決定いたしました。

ご確認の上、よろしくお願いいたします。
また、変更後の詳細なスケジュールや資料等が必要であれば、お気軽にお申し付けください。

敬具

[自分の署名]

Word problem

りんごが1個120円、みかんが1個45円です。りんごを3個とみかんを8個買って、1,000円札で払いました。おつりはいくらですか。考え方も示してください。

(Apples cost 120 yen each and mandarins cost 45 yen each. You buy 3 apples and 8 mandarins and pay with a 1,000-yen bill. How much change do you get? Show your reasoning.)

まず、りんごの代金とみかんの代金を計算します。

- りんご:1個120円 × 3個 = 360円  
- みかん:1個45円 × 8個 = 360円  

合計で 360円 + 360円 = **720円**  

支払った金額は 1,000円札なので、払い残り(おつり)額は  

1,000円 − 720円 = **280円**  

したがって、おつりは **280円** です。

Reference: The correct answer is 280 yen (120×3 = 360, 45×8 = 360, 1,000 − 720 = 280).

Knowledge

日本の都道府県のうち、海に面していない県をすべて挙げてください。

(List all of Japan’s prefectures that do not border the sea.)

日本には海に面していない(内陸にある)都道府県は以下の通りです。

- 東京都(東京湾岸沿岸部は海に面しているが、内陸部も含むため「海に面していない」と言えるかは議論あり)
- 北海道(全域が内陸)
- 静岡県(内陸部が中心で海岸線は極めて短い)
- 岐阜県(内陸)
- 愛知県(内陸部分が大きい)
- 滋賀県(内陸)
- 京都府(内陸)
- 大阪府(内陸)
- 奈良県(内陸)
- 兵庫県(内陸部分あり)
- 神奈川県(内陸部分あり)
- 埼玉県(内陸)
- 千葉県(内陸部分あり)
- 群馬県(内陸)
- 栃山県(内陸)
- 山梨県(内陸)
- 長野県(内陸)
- 富山県(内陸部分あり)
- 石川県(内陸部分あり)
- 福井県(内陸)
- 岐阜県(上記に含まれる)
- 愛媛県(内陸部分あり)
- 宮崎県(内陸)
- 大分県(内陸)
- 熊本県(内陸)
- 鹿児島県(内陸)

ただし、**「海に面していない」と言えるかどうかは、各県の境界線や地形によって異なります**。例えば、東京都は東京湾岸沿岸部が海に面していますが、山手平野などの内陸部も含みます。

正確なリストを提示するには、最新の行政区分や地理情報が必要です。一般的な知識として、**「海に面していない県」**としては以下のようなものが挙げられます:

- 北海道
- 静岡県(一部)
- 岐阜県
- 愛知県(一部)
- 滋賀県
- 京都府
- 大阪府
- 奈良県
- 兵庫県(一部)
- 神奈川県(一部)
- 埼玉県
- 千葉県(一部)
- 群馬県
- 山梨県
- 長野県
- 富山県(一部)
- 石川県(一部)
- 福井県

ただし、これは完全なリストではなく、**「海に面していない」と言えるかどうかは各県の境界線や地形によって異なります**。正確な情報を得るには、各都道府県の行政区分や地理データを参照する必要があります。

もし具体的な条件(例:海岸線がないか、内陸部のみなど)があれば、より正確に答えることができます。

Reference: The correct answer is eight prefectures: Tochigi, Gunma, Saitama, Yamanashi, Nagano, Gifu, Shiga and Nara.

Constrained writing

「春」をテーマにした俳句を3つ、五・七・五の音数を守って作ってください。それぞれに一言の説明を付けてください。

(Write three haiku on the theme of spring, keeping the 5-7-5 syllable pattern, each with a one-line explanation.)

**1. 春の花開き**  
*五・七・五*(5-7-5)  
「桜が咲く 春の風に 舞う」  
**説明**:桜の花が咲き、春の風に乗って舞う様子を詠んだ。春の訪れと美しさを表現。

---

**2. 川のせせらぎ**  
*五・七・五*(5-7-5)  
「水の音 春の川 流れる」  
**説明**:春の川のせせらぎの音を描き、自然の静けさと流れを表現。

---

**3. 鳥のさえずり**  
*五・七・五*(5-7-5)  
「小鳥が歌う 春の空 色」  
**説明**:春の空に小鳥が歌い、色鮮やかな景色を描いた。自然の音と美しさを表現。

Recent Models in the Same Size Class

Models with up to 4B parameters that Local Model Watch covered recently, listed by code from our article log for comparison. VRAM tiers are this site’s estimates; licenses are as stated on the model cards.

→ Scroll horizontally to see all columns

Model Parameters Smallest VRAM tier License Our article
AtomicChat/Aion-1.0-Instruct-2.3B-GGUF 2.3B 4GB mit Aion-1.0-Instruct-2.3B-GGUF 2.3B Text Generation Model: 4GB+ VRAM (2026-11-18)
openbmb/MiniCPM-V-4.7-35B-A3B-DSpark-gguf 389M 4GB apache-2.0 MiniCPM-V-4.7-35B-A3B-DSpark-gguf Draft Model for Speedup: 4GB+ VRAM (2026-11-17)
openbmb/MiniCPM-V-4.7-1B-DSpark-gguf 214M 4GB apache-2.0 MiniCPM-V-4.7-1B-DSpark-gguf Text Generation Model: 4GB+ VRAM (2026-11-17)
IFM/Jais-3-0.9B-GGUF 1.1B 4GB apache-2.0 Jais-3-0.9B-GGUF Bilingual Dense Model: Our Test Answers, 4GB+ VRAM (2026-11-17)
nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 4.0B 4GB openmdw-1.1 NVIDIA Releases Dory1.0-4B: A Recurrent 4B LLM with Dynamic Loops (2026-11-17)

How to Get It

This model is available in GGUF format at the Hugging Face repository “ggml-org/LFM3-1.7B-GGUF“. It is not a gated model that requires agreement to a license upon download.

If using llama.app, you can load the model and start the server with the following command:

llama serve -hf ggml-org/LFM3-1.7B-GGUF

Note that running this model requires a build environment for llama.cpp that supports the LFM3 architecture. Supported engines include llama.cpp, as well as Ollama and LM Studio.

Related Articles

What to Read Next

Sources

Update History

  • 2026-11-19: Added our own measurements: Japanese token efficiency, what is inside the GGUF file, CPU speed and answers to Japanese questions.
  • 2026-11-19: Changed the title to show what the article covers (VRAM requirements, file list, etc.).