selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

LFM3-1.7BvsQwen3-0.6B

Side-by-side specs, memory needs, licence, runtimes and benchmarks. Add up to 4 models.

--vram
any
--gpus
--quant
--add

On your GPU

  • LFM3-1.7B
  • Qwen3-0.6B
  • Qwen3-0.6B is the lightest to run (≈1.8 GB at its recommended quantisation).
  • LFM3-1.7B has the longest context (256K tokens).
  • Qwen3-0.6B has the most permissive licence (Apache 2.0: commercial use ok).
  • Qwen3-0.6B is the most downloaded in the last 30 days.
Property
Overview
TaskChatChat
Parameters1.7B752M
Active parametersdensedense
Context length256K40K
Input → output
Released23 Oct 202627 Apr 2025
Licencelfm1.0Apache 2.0
Gated downloadnono
Memory and speed
Fits your hardwarepick a GPU above≈1.9 GB≈1.8 GB
VRAM at Q4≈1.9 GB≈1.8 GB
VRAM at Q8≈2.7 GB≈2.2 GB
VRAM at FP16≈4.3 GB≈2.9 GB
Est. generation speedpick a GPU above——
Popularity
Downloads / 30 days100.6K18.5M
Likes1.8K1.9K
Benchmarks
BFCL v4—23.9%
Runs with
Jan
KoboldCpp
llama.cpp
LM Studio
LocalAI
MLX
Ollama
ONNX Runtime
Transformers
ExLlamaV2—
SGLang—
Text Generation Inference—
vLLM—
Ollama taglfm3:1.7bqwen3:0.6b
Architecture
ArchitectureLfm3ForCausalLMQwen3ForCausalLM
Layers2428
Attention / KV heads16 / 416 / 8
Vocabulary128 000151 936
Weights dtypebfloat16bfloat16
GGUF availableyesyes

Keep comparing

Memory figures are estimates for an 8K context at the chosen quantisation (Q4_K_M for LLMs by default). Speeds are rough, memory-bound estimates. Self-reported benchmark scores come from the model authors.