Run LFM3-1.7B on M4

LFM3-1.7B runs at 93 tok/s (92.5 as recorded) on a 16 GB M4, a community run reported by omlx.ai (MLX, 4-bit, 1k context). It needs about 3 GB of unified memory. It is Liquid AI's 1.73B dense model, released under the LFM Open licence.

Speed
community93
tok/s
First Token
community1.247
seconds
RAM Needed
3
GB minimum
Engine
MLX
of this figure
Want to run LFM3-1.7B faster — or step up to bigger models? Find your Mac →

The figure

This is a community run: a real run, reported by its source and linked below. The index records it as given and never adjusts it.

MetricValue
Tokens per secondcommunity93 tok/s
Provenancecommunity run · omlx.ai
Quantization4-bit
EngineMLX
Time to first token1.247s
Prompt processing (prefill)community810 tok/s at 4K context
Context1k
Figure dated2026-11-14
Minimum RAM3 GB
Parameters1.73B
Chip bandwidth120 GB/s

Every figure for LFM3-1.7B on the M4

Each row as the index holds it, with its source. “Mac RAM” is the memory of the Mac a run used, not the model’s requirement (3 GB). The headline figure follows the reference rule: the lower-median measured plain-decode run at 5 bits or fewer (a run whose quant was not stated counts only when no stated-quant run qualifies), else the estimate; a run marked “not used to anchor estimates” is listed here but never becomes the headline.

SpeedQuantEngineMac RAMContextSourceNote
community92.5 tok/s4-bitMLX16 GB1komlx.ai8-core GPU (120 GB/s, the same as the 10-core part). oMLX v0.8.0rc1, MTP/DFlash off; LiquidAI LFM3-1.7B-MLX-4bit; 1.5 GB peak. 87.9 tok/s at 4k. A second session the same day (r1gbupx9) read 91.7.
Provenance key
sourced
Published figure — vendor or press review
estimated
community
Linked individual run — blog, forum or submission

Speed by context depth: LFM3-1.7B on the M4

Measured figures only, each from its source’s own session; the index does not estimate long-context or prefill speed, so a missing cell is —. Each @ column is plain decode with that many tokens already in the prompt, from the same session as the row’s short-context figure: decode slows as the KV cache it reads grows. Prefill is prompt processing, the tokens per second a runtime reads a prompt at; with long agent contexts it sets how long you wait before the first token.

RunDecode (row)@4KPrefillTTFT
MLX · 4-bit · 16 GB Mac · omlx.aicommunity92.5 tok/s
1k
community87.9 tok/scommunity810 tok/s
at 4K
community1.247 s

Setup: run LFM3-1.7B on a Mac

Ollama — install it (brew install ollama or the app from ollama.com), then pull the model by its tag, checked on ollama.com on 2026-11-14:

ollama run lfm3:1.7b

MLX — pick a build from Hugging Face’s MLX listings, then run it with its repo id:

pip install -U mlx-lm
mlx_lm.generate --model <repo-id> --prompt "Hello"

Downloads: Ollama page → · Hugging Face model card →

The 93 tok/s figure above came from MLX at 4-bit; other runtimes and quants run at different speeds.

Prefer a GUI? LM Studio downloads GGUF and MLX builds in one click. Step by step: install Ollama on a Mac.

LFM3-1.7B on other Apple Silicon chips

ChipSpeedQuantEngineBandwidth
M5 Maxestimated285 tok/sQ4_K_Mllama.cpp614 GB/s
M5 Proestimated214 tok/sQ4_K_M—307 GB/s
M5community117 tok/s4-bitMLX153 GB/s
M1 Procommunity84 tok/s4-bitMLX200 GB/s

For the estimated figures here, the bandwidth formula’s typical error for dense models is ±9%, against 75 measured runs of 20 models (how it is computed).

About LFM3-1.7B

Liquid AI's 1.7B tri-hybrid model (27 Oct 2026, LFM Open License v1.0), ~1 GB at 4-bit, reasoning on by default. Card: SWE-bench Verified 43.4 (OpenCode), IFEval 96.7, GPQA Diamond 63.4. llama.cpp b11960+ and Ollama (lfm3:1.7b) run it, and on MLX it runs through mlx-vlm 0.7.11+; community oMLX runs measured ~117 tok/s on a base M5 and ~93 on an M4 (4-bit).

LFM Open licence · 1.73B · 256K context · LLMCheck capability 17/50 · see its leaderboard rank →

Memory by context length

Estimated, not measured: the index’s minimum for LFM3-1.7B (3 GB — the weights at the quant that fits, plus ~2 GB) plus the KV cache for that many tokens, computed from its config.json. hybrid short-conv: attention on 11 of 24 layers (5 sliding, window 256); conv layers keep a constant state. A 16-bit KV cache is the default in llama.cpp, Ollama, LM Studio and mlx-lm; a q8 cache halves the KV part. Coding agents hold 64K tokens or more. Runtimes also add a buffer that grows with context, so leave a few GB spare. macOS lets the GPU use only part of a Mac’s memory by default (figures for an M4 Mac on macOS 15, the lower default; macOS 26 allows more): “fits” means within the GPU memory a Mac of that tier can use with the limit raised, a tier tagged raise GPU limit fits only after sudo sysctl iogpu.wired_limit_mb raises it (the tooltip has the exact command), and the last column is the smallest Mac memory size whose default limit holds it (the GPU memory limit).

ContextKV cache (16-bit)RAM neededRAM, q8 KVSmallest tier that fitsWithin the default GPU limit
8K~0.1 GB~3.1 GB~3.0 GB8 GB Macsame
32K~0.4 GB~3.4 GB~3.2 GB8 GB Macsame
64K (coding-agent context)~0.8 GB~3.8 GB~3.4 GB8 GB Macsame
128K~1.6 GB~4.6 GB~3.8 GB8 GB Macsame

System requirements

Macs with the M4

🛒 Get a Mac that runs LFM3-1.7B

LFM3-1.7B needs about 3 GB of unified memory. These Apple Silicon Macs have enough to run it.

As an Amazon Associate, LLMCheck earns from qualifying purchases. These affiliate links cost you nothing extra and help keep the LLMCheck index free.

Compare More Models

See how LFM3-1.7B stacks up against other models on your specific Mac hardware.

Open Compare Tool Full Leaderboard