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-VL-450M

Hugging Face ↗ Follow changes

LiquidAI lfm1.0 out 7 months ago

449M parameters, images and text in, text out, 125K tokens of context.

Verified 2026-10-25 Rank #608 in LLMs · score 73.9 method v1 Sources 3 Downloads 30d 41.8k · #396 of 619 tracked

Can I run it?0.3GB at INT4Estimatefits a 16 GB cardmethod
Can I use it?lfm1.0Custom license; read the repo LICENSE before commercial hostinglicence file2026-10-24
Do people use it?41.8kdownloads, 30 days#396 of 619 we track · below the median+3 likes, downloads +6% this week233 likesHugging Face2026-10-24

Worth reading

Picked by rule from 10 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

10 best of 10 tracked

The 2 picks above, plus these 8. 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 · 11 HN pts · 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 · 1 HN pts
  7. ArticleLFM2.5-350M: No Size Left Behindliquid.ai · 2026-04-01 · official
  8. DiscussionLFM2.5-350Mtwitter.com · 2026-04-02

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: 33 quantized builds and 15 Spaces.

  • As a service0no API provider listed
  • On own hardware33quantized builds
  • Built on6035 fine-tunes · 25 adapters · 0 merges
  • In apps15Hugging Face Spaces

Top Spaces running it: AST-1320/Qwen-2511-Plus-Flux.2-klein ♥ 238 · LiquidAI/LFM2.5-VL-450M-WebGPU ♥ 62 · tonywu71/neomme-retriever-demo ♥ 8 · prithivMLmods/Multimodal-Edge-Node ♥ 6

Will it fit?

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

Fits a 16 GB card at INT4

About 0.3 GB at INT4, 0.5 GB at FP8, 1.1 GB at BF16: weights only, so leave room for the KV cache.

16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.3 GB of weights (estimate; KV cache and activations extra)INT40.3 GBFP8: about 0.5 GB of weights (estimate; KV cache and activations extra)FP80.5 GBBF16: about 1.1 GB of weights (estimate; KV cache and activations extra)BF161.1 GB
16 GB card64 GB Mac2×80 GB8×80 GBINT4: about 0.3 GB of weights (estimate; KV cache and activations extra)INT40.3 GBFP8: about 0.5 GB of weights (estimate; KV cache and activations extra)FP80.5 GBBF16: about 1.1 GB of weights (estimate; KV cache and activations extra)BF161.1 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.

5 dated articles and 4 repo commits so far

Teal is the weights (2026-04-08), amber triangles are 5 dated articles (each links to it), grey ticks are repo commits.

MayJunJulAugSepOct2026-04-08: 4 repo commitsLFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge · 2026-08-12LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacB · 2026-08-21LFM2.5-Encoders for Fast Long-Context Inference on CPU · 2026-08-06Accomplishing More with LFM2.5-VL-DSpark · 2026-09-24LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud · 2026-04-27Weights 2026-04-08 · 2026-04-08Weights 2026-04-08now
MayJunJulAugSepOct2026-04-08: 4 repo commitsLFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge · 2026-08-12LFM2.5-DSpark: Up to 3.2x Faster Inference from H100 to MacB · 2026-08-21LFM2.5-Encoders for Fast Long-Context Inference on CPU · 2026-08-06Accomplishing More with LFM2.5-VL-DSpark · 2026-09-24LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud · 2026-04-27Weights 2026-04-08 · 2026-04-08Weights 2026-04-08now

What people built on it

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

93 derivatives on the Hub, led by 35 fine-tunes

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

33 quantized (Hugging Face model tree)35 fine-tunes (Hugging Face model tree)25 adapters (Hugging Face model tree)33 quantized35 fine-tunes25 adapters
33 quantized (Hugging Face model tree)35 fine-tunes (Hugging Face model tree)25 adapters (Hugging Face model tree)33 quantized35 fine-tunes25 adapters
GGUF LiquidAI/LFM2.5-VL-450M-GGUF for llama.cpp, Ollama, LM Studio 140.9k downloads MLX mlx-community/LFM2.5-VL-450M-6bit for Apple silicon 9.6k downloads

All variants →

Reference

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

Benchmarks

13 rows

Benchmarks 13 rows

measured third-party leaderboard rowauthor-reported number from the model card or eval files (13)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 BLINK · Understand images and video43.92—model author2026-10-07
author-reported IFEval · Chat and follow instructions45—model author2026-10-07
author-reported InfoVQA · Extract text from documents43.02—model author2026-10-07
author-reported MMBench · Understand images and video60.91—model author2026-10-07
author-reported MMMB68.09—model author2026-10-07
author-reported MMMU · Understand images and video32.67—model author2026-10-07
author-reported MMStar · Understand images and video43—model author2026-10-07
author-reported MMVet · Understand images and video41.1—model author2026-10-07
author-reported OCRBench · Extract text from documents684—model author2026-10-07
author-reported POPE86.93—model author2026-10-07
author-reported RealWorldQA · Understand images and video58.43—model author2026-10-07
author-reported RefCOCO · Understand images and video81.28—model author2026-10-07
author-reported Terminal-Bench · Write code with agents73.31—model author2026-10-07

Use-case evidence

7 use cases

What it is good at 7 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 tagimage-text-to-text
Librarytransformers
ArchitecturesLfm2VlForConditionalGeneration
Model typelfm2_vl
Dtypebfloat16
Params by dtypeBF16: 448,718,848
Size bandS
MoEno
Context length128,000 (config.json max_position_embeddings)
Languagesen, ja, ko, fr, es, de, ar, zh, pt
PapersarXiv:2511.23404
Licenselfm1.0 (other)
Hub snapshot2026-10-25

Servability

vLLM · SGLang

Servability

vLLM: supported · Lfm2VlForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Lfm2VlForConditionalGeneration · 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-VL-450M: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.

BF16 (2 bytes/param)1.1 GB · Edge (<4 GB)
FP8 (1 byte/param)0.5 GB · Edge (<4 GB)
INT4 (0.5 bytes/param)0.3 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/LFM2.5-350M (same organization) Family: LiquidAI/LFM2.5 93 derivatives on the Hub

Base model: LiquidAI/LFM2.5-350M

quantized: 33finetune: 35adapter: 25

Top variants (40 shown, ranked by downloads)

Relation
Quant
40 rows
Variant Relation Quant Params Band License Author Likes Downloads
LiquidAI/LFM2.5-VL-450M-GGUF quantized GGUF — — other LiquidAI 72 140.9k
LiquidAI/LFM2.5-VL-450M-ONNX quantized — — — other LiquidAI 33 16.8k
LiquidAI/LFM2.5-VL-450M-Extract finetune — 449M S other LiquidAI 57 7.1k
mlx-community/LFM2.5-VL-450M-6bit quantized — 449M S other mlx-community 0 9.6k
LiquidAI/LFM2.5-VL-450M-MLX-4bit quantized MLX 449M S other LiquidAI 10 4.6k
erikku-sama/lfm2-electronics-vlm-v2 quantized GGUF 449M S other erikku-sama 8 3.4k
onnx-community/LFM2.5-VL-450M-ONNX quantized — — — other onnx-community 2 2.9k
LiquidAI/LFM2.5-VL-450M-MLX-bf16 finetune MLX 449M S other LiquidAI 11 2.4k
litert-community/LFM2.5-VL-450M quantized — — — other litert-community 3 2.2k
LiquidAI/LFM2.5-VL-450M-MLX-8bit quantized MLX 449M S other LiquidAI 10 1.7k
NobodyWho/LFM2.5-VL-450M-GGUF quantized GGUF — — other NobodyWho 0 1.9k
LiquidAI/LFM2.5-VL-450M-MLX-6bit quantized MLX 449M S other LiquidAI 9 1.2k
5ch4um1/lfm2.5-vrsbench-lora-450m quantized GGUF 449M S — 5ch4um1 1 1.6k
5ch4um1/lfm2.5-vrsbench-EUROSAT-terrain-lora-450m quantized GGUF 449M S — 5ch4um1 0 1.5k
LiquidAI/LFM2.5-VL-450M-MLX-5bit quantized MLX 449M S other LiquidAI 8 808
5ch4um1/lfm2.5-vrsbench-MADOS-maritime-lora-450m quantized GGUF 449M S — 5ch4um1 0 832
samwell/galamsey-v9-e3-onnx quantized — — — other samwell 2 716
lewiswatson/Frame2KG-LFM-2.5-450m-JSON quantized GGUF — — — lewiswatson 0 711
mlx-community/LFM2.5-VL-450M-8bit quantized -8bit 449M S other mlx-community 0 696
opencal/opencal-base quantized — — — apache-2.0 opencal 0 648
webbrain-one/webbrain-vl-2-450M finetune — 449M S other webbrain-one 5 394
ermiaazarkhalili/LFM2.5-VL-450M-SFT-Fable5-Glint adapter — 449M S other ermiaazarkhalili 0 623
urbanspr1nter/lfm2.5vl-450m-deforestation-classifier quantized GGUF 449M S other urbanspr1nter 0 519
jpmarindiaz/lfm2-flood quantized GGUF 516M S apache-2.0 jpmarindiaz 0 471
Arty1001/aquaveritas-lfm-GGUF quantized GGUF — — apache-2.0 Arty1001 0 401
ermiaazarkhalili/LFM2.5-VL-450M-SFT-Fable5-Glint-GGUF adapter GGUF — — other ermiaazarkhalili 0 383
sanskar003/MakeModel-VLM-450M finetune — 449M S other sanskar003 5 118
Varun12387/LFM2.5-VL-450M-SFT-GGUF quantized GGUF 449M S other Varun12387 0 363
NuTonic/lspace finetune — 449M S apache-2.0 NuTonic 2 212
lewiswatson/Frame2KG-LFM-2.5-450m-CT quantized GGUF — — — lewiswatson 0 305
akashreddy2103/landfill adapter — — — apache-2.0 akashreddy2103 0 297
Shoozes/lfm2.5-450m-vl-orbit-satellite quantized GGUF — — mit Shoozes 0 285
harishforaiandml/LFM2.5-VL-450M-Q2_K-GGUF quantized Q2_K — — other harishforaiandml 0 282
munish0838/crop-burn-detector finetune — 449M S cc-by-4.0 munish0838 1 204
marcelo-earth/LFM2.5-VL-450M-satellite-triage adapter — 516M S apache-2.0 marcelo-earth 1 203
Dadm-n/LFM2.5-VL-450M-mlx finetune mlx 449M S other Dadm-n 0 221
WobblyDopamine/SatDiff-LFM2.5-VL-450M-stage1 adapter GGUF 449M S other WobblyDopamine 0 213
nathanrchn/coffee-leaf-lfm2.5-vl-450m-GGUF quantized GGUF — — other nathanrchn 0 201
johnny711/argus-lfm-lora adapter — — — apache-2.0 johnny711 4 0
Alfaxad/Vector-LFM2.5-VL-450M adapter — 449M S other Alfaxad 0 197

Repository changelog

4 commits

Repository changelog 4 commits

DateCommitSHA
2026-04-08 Upload README.md with huggingface_hub fc6221c
2026-04-08 Upload README.md with huggingface_hub 9125647
2026-04-08 Upload folder using huggingface_hub a8f47c5
2026-04-08 initial commit e8c6013

Articles and coverage

10 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.

Official4

LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge
liquid.ai2026-08-12officialabout the LFM2.5 familyHN 11 pts · 3 comments2,057 words
Liquid AI releases LFM2.5-VL-3B, a non-reasoning vision-language model that integrates a SigLIP2 400M NaFlex encoder and doubles its vocabulary to 128K. It pre-trains on ~34T tokens and decodes 228 tokens/s on an Apple M5 Max. The model achieves about 11K tokens per second output throughput on a single NVIDIA H100.written, reviewed, cited
  • The model doubles its vocabulary to 128K by extending the existing tokenizer in place to support non-Latin scripts.
  • LFM2.5-VL-3B runs on a phone, decoding 20 tokens/s on a Galaxy S26 Ultra for private on-device execution.
  • Evaluation uses vLLM 0.26.0 and non-reasoning mode where models are prompted to directly answer without reasoning.
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-VL-450M: Structured Visual Intelligence, Edge to Cloud
liquid.ai2026-04-27officialHN 2 pts · 0 comments956 words

Analysis1

Accomplishing More with LFM2.5-VL-DSpark
2tinteractive.com2026-09-24about the LFM2.5 familyHN 1 pts · 1 comments888 words
LiquidAI released LFM2.5-VL-DSpark on September 24, 2026, which accelerates vision-language model training from 20 days to 3 days. The architecture enhances computational efficiency by 20 percent and includes new training data sets curated by experts. It features integration APIs and GDPR compliance details available on Hugging Face.written, reviewed, cited
  • The model includes a new neural architecture designed to minimize latency for real-time video analysis.
  • The architecture is more energy-efficient, reducing the computational resources and costs associated with training.
All coverage tracked for this model (10) — append-only, nothing is ever removed
TitleSourcePublishedFirst seenSignals
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

  • GOT-OCR-2.0-hf · 561M
    same size band (S), same task (image-text-to-text), permissive license
  • surya-ocr-2 · 686M
    same size band (S), same task (image-text-to-text), more providers
  • GOT-OCR2_0 · 716M
    same size band (S), same task (image-text-to-text), permissive license

Sources

3 sources

Sources

Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:41 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:13 UTC · Last verified 2026-10-25

Follow LFM2.5-VL-450M

Watch this model

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