Model Follower.

Up to date

Rebuilt
2026-10-25 06:54 UTC
Release radar
117 on watch
Hub snapshot
2026-10-25
Re-checked
daily · every number sourced and dated

Models › LFM2.5 family

LFM2.5-1.2B-Instruct

Hugging Face ↗ Follow changes

LiquidAI lfm1.0 out 10 months ago

1.17B parameters, text in, text out, 125K tokens of context.

Verified 2026-10-25 Rank #284 in LLMs · score 84.4 method v1 Sources 3 Downloads 30d 112k · #309 of 619 tracked

Changed this week: 6 new quantized builds on the Hub (126 in all).2026-10-24

Can I run it?0.7GB at INT4Estimatefits a 16 GB cardmethodCommunity: Liquid AI LFM2.5 1.2B: On-Device Agentic AI, Under 1GB | Toolbit.ai →
Can I use it?lfm1.0Custom license; read the repo LICENSE before commercial hostinglicence file2026-10-24
Do people use it?112kdownloads, 30 days#309 of 619 we track · top 50%+2 likes, downloads −3% this week684 likesHugging Face2026-10-24

Worth reading

Picked by rule from 16 tracked pieces: the lab's own post, the most shared, a field report, the latest.

Latest · 2 months ago

LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacB

Liquid AI releases DSpark draft models for LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B, integrating speculative decoding via llama.cpp and SGLang.

  • The 8B-A1B draft model adds 65.5M parameters for its Markov head compared to 33.6M in the 1.2B and 2.6B variants.
  • For the 1.2B-Instruct draft, validation loss consistently tracks acceptance rate improvements across all training epochs.
liquid.ai 2026-08-21 18 HN pts 1.9k words about the LFM2.5 family family

16 best of 16 tracked

The 4 picks above, plus these 12. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.

  1. ArticleLFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edgeliquid.ai · 2026-08-12 · official
  2. ArticleLFM2.5-8B-A1B: An Even Better On-Device Mixture-of-Expertsliquid.ai · 2026-05-28 · official
  3. ArticleLFM2.5-2.6B: On-Device Agentsdocs.liquid.ai · 2026-08-04 · official
  4. ArticleLFM2.5-Encoders for Fast Long-Context Inference on CPUhuggingface.co · 2026-08-06 · 1 HN pts
  5. ArticleLFM2.5-2.6B: Deploy Agents Everywhereliquid.ai · 2026-08-04 · official
  6. ArticleAccomplishing More with LFM2.5-VL-DSpark2tinteractive.com · 2026-09-24
  7. ArticleLFM2.5-350M: No Size Left Behindliquid.ai · 2026-04-01 · official
  8. ArticleLFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloudliquid.ai · 2026-04-27 · official
  9. DiscussionLFM2.5-350Mtwitter.com · 2026-04-02
  10. ArticleLFM2.5 1.2B Instruct – 32k context, open weights | LLM Referencellmreference.com · undated
  11. ArticleLFM2.5-1.2B-Instruct - Dataconomydataconomy.com · undated
  12. ArticleLiquidAI/LFM2.5-1.2B-Instruct: 2.6 GB VRAMaquanode.io · undated

Where it's used

Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.

No API provider serves it, so its downloads are people running it themselves: 126 quantized builds and 30 Spaces.

  • As a service0no API provider listed
  • On own hardware126quantized builds
  • Built on260148 fine-tunes · 111 adapters · 1 merges
  • In apps30Hugging Face Spaces

Top Spaces running it: AlexWortega/activation-oracle ♥ 16 · hadadxyz/LiquidAI-LFM2.5-1.2B-Instruct ♥ 11 · stanley-00/slm-testing ♥ 9 · FlameF0X/lfm2 ♥ 7

Will it fit?

Weights only, an estimate: leave room for the KV cache. Method

Fits a 16 GB card at INT4

About 0.7 GB at INT4, 1.4 GB at FP8, 2.8 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.7 GB of weights (estimate; KV cache and activations extra)INT40.7 GBFP8: about 1.4 GB of weights (estimate; KV cache and activations extra)FP81.4 GBBF16: about 2.8 GB of weights (estimate; KV cache and activations extra)BF162.8 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.7 GB of weights (estimate; KV cache and activations extra)INT40.7 GBFP8: about 1.4 GB of weights (estimate; KV cache and activations extra)FP81.4 GBBF16: about 2.8 GB of weights (estimate; KV cache and activations extra)BF162.8 GB

Estimate params × bytes per param × 1.2 · how we estimate

The story so far

Weights, every dated article, and repo commits on one rail.

2 dated articles and 19 repo commits so far

Teal is the weights (2026-01-06), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.

FebMarAprMayJunJulAugSepOct2026-08-24: 1 repo commit2026-08-05: 1 repo commit2026-08-03: 1 repo commit2026-06-16: 1 repo commit2026-03-30: 2 repo commits2026-03-28: 1 repo commit2026-02-24: 1 repo commit2026-02-02: 1 repo commit2026-01-27: 1 repo commit2026-01-26: 1 repo commit2026-01-22: 1 repo commit2026-01-20: 1 repo commit2026-01-19: 1 repo commit2026-01-09: 2 repo commits2026-01-07: 1 repo commit2026-01-06: 2 repo commitsLFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacB · 2026-08-21LFM2.5-Encoders for Fast Long-Context Inference on CPU · 2026-08-06Weights 2026-01-06 · 2026-01-06Weights 2026-01-06now
FebMarAprMayJunJulAugSepOct2026-08-24: 1 repo commit2026-08-05: 1 repo commit2026-08-03: 1 repo commit2026-06-16: 1 repo commit2026-03-30: 2 repo commits2026-03-28: 1 repo commit2026-02-24: 1 repo commit2026-02-02: 1 repo commit2026-01-27: 1 repo commit2026-01-26: 1 repo commit2026-01-22: 1 repo commit2026-01-20: 1 repo commit2026-01-19: 1 repo commit2026-01-09: 2 repo commits2026-01-07: 1 repo commit2026-01-06: 2 repo commitsLFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacB · 2026-08-21LFM2.5-Encoders for Fast Long-Context Inference on CPU · 2026-08-06Weights 2026-01-06 · 2026-01-06Weights 2026-01-06now

What people built on it

386 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.

386 derivatives on the Hub, led by 148 fine-tunes

From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.

126 quantized (Hugging Face model tree)148 fine-tunes (Hugging Face model tree)1 merge (Hugging Face model tree)111 adapters (Hugging Face model tree)126 quantized148 fine-tunes1 merge111 adapters
126 quantized (Hugging Face model tree)148 fine-tunes (Hugging Face model tree)1 merge (Hugging Face model tree)111 adapters (Hugging Face model tree)126 quantized148 fine-tunes1 merge111 adapters
GGUF LiquidAI/LFM2.5-1.2B-Instruct-GGUF for llama.cpp, Ollama, LM Studio 1.9M downloads MLX lmstudio-community/LFM2.5-1.2B-Instruct-MLX-8bit for Apple silicon 2M downloads AWQ kaitchup/LFM2.5-1.2B-Instruct-awq-asym for vLLM, SGLang 2.1k downloads

All variants →

Reference

Everything else, folded: the full tables, every source.

Benchmarks

1 row

Benchmarks 1 row

measured third-party leaderboard rowauthor-reported number from the model card or eval files (1)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
BenchmarkValueSetting / notesReported byDate
author-reported diamond (Idavidrein/gpqa)38.89—Model Card2026-01-27

What the card says

1 cited sentence

Summary generated, cited

LiquidAI/LFM2.5-1.2B-Instruct is a 1.17B parameter instruction-tuned model designed for on-device deployment.[1][2]

  1. parameters human
  2. “designed for on-device deployment”

Generated 2026-10-07 15:08 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.

Use-case evidence

4 use cases

What it is good at 4 use cases

Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.

Facts

22 fields

Listed in LLMs

Facts

Pipeline tagtext-generation
Librarytransformers
ArchitecturesLfm2ForCausalLM
Model typelfm2
Dtypebfloat16
Params by dtypeBF16: 1,170,340,608
Size bandS
MoEno
Context length128,000 (config.json max_position_embeddings)
Languagesen, ar, zh, fr, de, ja, ko, es
PapersarXiv:2511.23404
Licenselfm1.0 (other)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Lfm2ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Lfm2ForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed

vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.

Hardware fit

3 estimates

Hardware fit ESTIMATE

Weights-only memory for LFM2.5-1.2B-Instruct: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)2.8 GB · Edge (<4 GB)
FP8 (1 byte/param)1.4 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)0.7 GB · Edge (<4 GB)

Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.

All variants

40 variants

Lineage

Fine-tuned from LiquidAI/LFM2.5-1.2B-Base (same organization) Family: LiquidAI/LFM2.5 386 derivatives on the Hub

Base model: LiquidAI/LFM2.5-1.2B-Base

quantized: 126finetune: 148adapter: 111merge: 1

Top variants (40 shown, ranked by downloads)

Relation
Quant
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
lmstudio-community/LFM2.5-1.2B-Instruct-MLX-8bit quantized MLX 1.17B S other lmstudio-community 3 2M
lmstudio-community/LFM2.5-1.2B-Instruct-MLX-4bit quantized MLX 1.17B S other lmstudio-community 2 2M
lmstudio-community/LFM2.5-1.2B-Instruct-MLX-6bit quantized MLX 1.17B S other lmstudio-community 3 2M
LiquidAI/LFM2.5-1.2B-Instruct-GGUF quantized GGUF — — other LiquidAI 231 1.9M
farbodtavakkoli/OTel-LLM-1.2B-IT finetune — — — apache-2.0 farbodtavakkoli 2 1.6M
unsloth/LFM2.5-1.2B-Instruct-GGUF quantized GGUF — — other unsloth 46 212.3k
mlx-community/LFM2.5-1.2B-Instruct-4bit quantized -4bit 1.17B S other mlx-community 3 169k
lmstudio-community/LFM2.5-1.2B-Instruct-GGUF quantized GGUF — — other lmstudio-community 3 166k
NovachronoAI/LFM2.5-1.2B-Nova-Function-Calling-GGUF quantized GGUF — — apache-2.0 NovachronoAI 11 127k
unsloth/LFM2.5-1.2B-Instruct finetune — 1.17B S other unsloth 8 118.8k
LiquidAI/LFM2.5-1.2B-Instruct-ONNX quantized — — — other LiquidAI 37 19k
LiquidAI/LFM2.5-1.2B-Instruct-MLX-4bit quantized MLX 1.17B S other LiquidAI 17 19.2k
bartowski/LiquidAI_LFM2.5-1.2B-Instruct-GGUF quantized GGUF — — other bartowski 6 19.2k
LiquidAI/LFM2.5-1.2B-Instruct-MLX-8bit quantized MLX 1.17B S other LiquidAI 26 14.2k
litert-community/LFM2.5-1.2B-Instruct quantized — — — other litert-community 4 7.7k
Indexnusrefather/Super-Slop-Machina-Roleplay-1.2b finetune GGUF 1.17B S other Indexnusrefather 3 5.6k
LiquidAI/LFM2.5-1.2B-Instruct-DSpark finetune — 296M S other LiquidAI 35 3.4k
mradermacher/LFM2.5-1.2B-Instruct-i1-GGUF quantized GGUF — — other mradermacher 2 4.5k
LiquidAI/LFM2.5-1.2B-Instruct-MLX-bf16 finetune MLX 1.17B S other LiquidAI 14 3.7k
mlx-community/LFM2.5-1.2B-Instruct-8bit quantized -8bit 1.17B S other mlx-community 1 4k
huihui-ai/Huihui-LFM2.5-1.2B-Instruct-abliterated finetune — 1.17B S other huihui-ai 17 3.1k
ermiaazarkhalili/LFM2.5-1.2B-SFT-Claude-Opus-Reasoning-Unsloth-GGUF adapter GGUF — — other ermiaazarkhalili 1 3.5k
DavidAU/LFM2.5-1.2B-Instruct-Thinking-Claude-High-Reasoning finetune — 1.17B S — DavidAU 19 2.5k
NobodyWho/LFM2.5-1.2B-Instruct-GGUF quantized GGUF — — other NobodyWho 2 3.3k
VAGOsolutions/SauerkrautLM-Translator-LFM2.5-1.2B finetune — 1.17B S other VAGOsolutions 28 1.7k
Indexnusrefather/Super-Slop-Machina-Roleplay-1.2b-V4-ERP-Tolerant-Preview finetune GGUF 1.17B S other Indexnusrefather 0 2.9k
tatsuyaaaaaaa/LFM2.5-1.2B-Instruct-gguf quantized gguf — — — tatsuyaaaaaaa 0 2.8k
yasserrmd/GLM4.7-Distill-LFM2.5-1.2B finetune — 1.17B S apache-2.0 yasserrmd 11 1.9k
ermiaazarkhalili/LFM2.5-1.2B-SFT-Claude-Opus-Reasoning-Unsloth adapter — 1.17B S other ermiaazarkhalili 0 2.3k
MuXodious/LFM2.5-1.2B-Instruct-absolute-heresy finetune — 1.17B S other MuXodious 0 2.3k
LiquidAI/LFM2.5-1.2B-Instruct-MLX-6bit quantized MLX 1.17B S other LiquidAI 11 1.7k
Indexnusrefather/Super-Slop-Machina-Roleplay-1.2b-ERP-Tolerant-V2 adapter GGUF 1.17B S other Indexnusrefather 2 2.2k
d4rkninja/tanpo-marketing finetune — 1.17B S other d4rkninja 2 2.2k
mlboydaisuke/LFM2.5-1.2B-CoreAI quantized — — — other mlboydaisuke 0 2.2k
Indexnusrefather/LFM-2.5-1.2b-Instruct-roleplay-tuned adapter GGUF 1.17B S other Indexnusrefather 0 2.2k
YoussefElsafi/PlayerAI-1.2B-v1.5-GGUF quantized GGUF — — apache-2.0 YoussefElsafi 2 2.1k
kaitchup/LFM2.5-1.2B-Instruct-awq-asym quantized awq 1.32B S apache-2.0 kaitchup 1 2.1k
mradermacher/LFM2.5-1.2B-Instruct-GGUF quantized GGUF — — other mradermacher 0 2.1k
NovachronoAI/LFM2.5-1.2B-Nova-Function-Calling finetune — 1.17B S apache-2.0 NovachronoAI 5 1.8k
Indexnusrefather/LFM-2.5-1.2b-Instruct-roleplay-tuned-v2 adapter GGUF 1.17B S other Indexnusrefather 0 2.1k

Repository changelog

19 commits
changed in the last 7 daysOther: 2026-10-24all changes

Repository changelog 19 commits

DateCommitSHA
2026-08-24 Add DSpark 0f604ad
2026-08-05 Fix chat template: render assistant tool_calls (#20) df58c17
2026-08-03 Update README.md 3d2b14b
2026-06-16 Update LICENSE 868df74
2026-03-30 Fix Try LFM link to use playground.liquid.ai homepage 6314d2b
2026-03-30 Add Discord community link to README 97fd4c6
2026-03-28 Update README.md e7114d1
2026-02-24 Update README.md e129ea4
2026-02-02 add GRPO TRL fine-tuning notebook 719098d
2026-01-27 Add GPQA evaluation result (#9) 8c1348a
2026-01-26 Update README.md 8dc4e68
2026-01-22 Update README.md 47c527b
2026-01-20 Update README.md c20ca41
2026-01-19 Update README.md 43f1e80
2026-01-09 Update README.md 286966c
2026-01-09 Upload README.md with huggingface_hub 4393ab8
2026-01-07 Update README.md a4ba0a6
2026-01-06 Add files using upload-large-folder tool 205939a
2026-01-06 initial commit f3a7622

Articles and coverage

16 pieces

Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.

Official3

LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacB
liquid.ai2026-08-21officialabout the LFM2.5 familyHN 18 pts · 0 comments1,933 words
Liquid AI releases DSpark draft models for LFM2.5-1.2B-Instruct, LFM2.5-2.6B, and LFM2.5-8B-A1B, integrating speculative decoding via llama.cpp and SGLang. The draft models use 5 layers with a block size of 9 and total approximately ~300M parameters, achieving up to 3.18x GPU speedup and up to 2.87x on-device gains for the 2.6B variant. While the 1.2B and 2.6B models see consistent improvements, the 8B-A1B variant only achieves an 18% improvement on edge devices due to current MoE implementation constraints in llama.cpp's Metal backend.written, reviewed, cited
  • The 8B-A1B draft model adds 65.5M parameters for its Markov head compared to 33.6M in the 1.2B and 2.6B variants.
  • For the 1.2B-Instruct draft, validation loss consistently tracks acceptance rate improvements across all training epochs.
  • The 8B-A1B draft model achieves only an 18% improvement on edge devices because verifying k tokens activates more experts and weight traffic.
LFM2.5-Encoders for Fast Long-Context Inference on CPU
huggingface.co2026-08-06authorityabout the LFM2.5 familyHN 1 pts · 0 comments1,103 words
Liquid AI releases LFM2.5-Encoder-230M and LFM2.5-Encoder-350M, two new models designed for fast long-context inference on CPU. These encoders support an 8,192-token context and are about 3.7× faster than ModernBERT-base at long context lengths. The models are trained with a masked language objective that masks 30% of tokens during training to achieve strong performance on GLUE, SuperGLUE, and multilingual tasks.written, reviewed, cited
  • The 350M variant is the choice when accuracy matters most.
  • The 230M variant is the choice for tighter hardware or higher throughput.
  • A base encoder provides general-purpose representations rather than direct task outputs.
LFM2.5-1.2B-Instruct - Liquid Docs
docs.liquid.aiofficial

Analysis5

LFM2.5-1.2B-Instruct - Intelligence, Performance & Price Analysis | Artificial Analysis
artificialanalysis.aiauthority599 words
LFM2.5 1.2B Instruct – 32k context, open weights | LLM Reference
llmreference.com
Liquid AI LFM2.5 1.2B: On-Device Agentic AI, Under 1GB | Toolbit.ai
toolbit.ai
LFM2.5-1.2B-Instruct - Dataconomy
dataconomy.com
LiquidAI/LFM2.5-1.2B-Instruct: 2.6 GB VRAM
aquanode.io
All coverage tracked for this model (16) — append-only, nothing is ever removed
TitleSourcePublishedFirst seenSignals
LFM2.5-1.2B-Instruct - Liquid Docsdocs.liquid.ai official—2026-10-14exa
LFM2.5-1.2B-Instruct - Intelligence, Performance & Price Analysis | Artificial Analysisartificialanalysis.ai—2026-10-14exa
LFM2.5 1.2B Instruct – 32k context, open weights | LLM Referencellmreference.com—2026-10-14exa
Liquid AI LFM2.5 1.2B: On-Device Agentic AI, Under 1GB | Toolbit.aitoolbit.ai—2026-10-14exa
LFM2.5-1.2B-Instruct - Dataconomydataconomy.com—2026-10-14exa
LiquidAI/LFM2.5-1.2B-Instruct: 2.6 GB VRAMaquanode.io—2026-10-14exa
Accomplishing More with LFM2.5-VL-DSpark2tinteractive.com2026-09-242026-10-07hn
LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacBliquid.ai official2026-08-212026-10-07hn
LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edgeliquid.ai official2026-08-122026-10-07hn
LFM2.5-Encoders for Fast Long-Context Inference on CPUhuggingface.co2026-08-062026-10-07hn
LFM2.5-2.6B: On-Device Agentsdocs.liquid.ai official2026-08-042026-10-07hn
LFM2.5-2.6B: Deploy Agents Everywhereliquid.ai official2026-08-042026-10-07hn
LFM2.5-8B-A1B: An Even Better On-Device Mixture-of-Expertsliquid.ai official2026-05-282026-10-07hn
LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloudliquid.ai official2026-04-272026-10-07hn
LFM2.5-350Mtwitter.com2026-04-022026-10-07hn
LFM2.5-350M: No Size Left Behindliquid.ai official2026-04-012026-10-07hn

Papers

1 paper

Papers 1

arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).

Similar models

3 models

Similar models

  • Llama-3.2-1B-Instruct · 1.24B
    same size band (S), same task (text-generation), more providers
  • Llama-3.2-1B · 1.24B
    same size band (S), same task (text-generation), more providers
  • TinyLlama-1.1B-Chat-v1.0 · 1.10B
    same size band (S), same task (text-generation), more providers, permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:38 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:08 UTC · Last verified 2026-10-25

Follow LFM2.5-1.2B-Instruct

Watch this model

Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.