Skip to content

MODEL-MATRIX.md

Latest commit

 

History

History
75 lines (63 loc) · 4.76 KB

MODEL-MATRIX.md

File metadata and controls

75 lines (63 loc) · 4.76 KB

Model Implementation Matrix

Maintainer reference only. This file is excluded from the Mintlify docs build via .mintignore. Last updated: 2026-02-06

Quick Reference: Snippet Mapping

Category Snippet files Used by
Standard text snippets/quickstart/text-transformers.mdx, text-vllm.mdx, text-llamacpp.mdx LFM2.5-1.2B-Instruct, Thinking, JP; LFM2-8B-A1B, 2.6B, 2.6B-Exp, 1.2B, 700M, 350M
Vision snippets/quickstart/vl-transformers.mdx, vl-vllm.mdx, vl-llamacpp.mdx LFM2.5-VL-1.6B; LFM2-VL-3B, VL-1.6B, VL-450M
Audio Inline (unique code) LFM2.5-Audio-1.5B, LFM2-Audio-1.5B
ColBERT Inline (PyLate) LFM2-ColBERT-350M
Nanos Inline (custom prompts/temps) Extract, RAG, Transcript, Math, ENJP-MT, PII-Extract-JP

Full Implementation Matrix

Model HF ID Transformers llama.cpp vLLM Ollama MLX ONNX Notes
LFM2.5 Text
1.2B-Instruct LiquidAI/LFM2.5-1.2B-Instruct Standard Standard Standard Standard Yes Yes Flagship
1.2B-Thinking LiquidAI/LFM2.5-1.2B-Thinking Standard Standard Standard Standard Yes Yes
1.2B-Base LiquidAI/LFM2.5-1.2B-Base Base (no chat) -- Base (no chat) -- -- Yes Raw text completion
1.2B-JP LiquidAI/LFM2.5-1.2B-JP Standard Standard Standard Standard Yes Yes Japanese
LFM2.5 Vision & Audio
VL-1.6B LiquidAI/LFM2.5-VL-1.6B VL (v5 req) VL (--image) VL (custom wheel) NO (PR#14069) Yes Yes
Audio-1.5B LiquidAI/LFM2.5-Audio-1.5B liquid-audio Custom binary NO NO NO Yes
LFM2 Text
8B-A1B LiquidAI/LFM2-8B-A1B Standard Standard Standard Standard Yes Yes MoE
2.6B LiquidAI/LFM2-2.6B Standard Standard Standard Standard Yes Yes
2.6B-Exp LiquidAI/LFM2-2.6B-Exp Standard Standard Standard Standard NO NO
1.2B LiquidAI/LFM2-1.2B Standard Standard Standard Standard Yes Yes Deprecated
700M LiquidAI/LFM2-700M Standard Standard Standard Standard Yes Yes
350M LiquidAI/LFM2-350M Standard Standard Standard Standard Yes Yes
LFM2 Vision
VL-3B LiquidAI/LFM2-VL-3B VL (v5 req) VL VL (custom wheel) NO (PR#14069) Yes Yes
VL-1.6B LiquidAI/LFM2-VL-1.6B VL (v5 req) VL VL (custom wheel) NO (PR#14069) Yes Yes Deprecated
VL-450M LiquidAI/LFM2-VL-450M VL (v5 req) VL VL (custom wheel) NO (PR#14069) Yes Yes
LFM2 Audio
Audio-1.5B LiquidAI/LFM2-Audio-1.5B liquid-audio Custom binary NO NO NO NO Deprecated
Nanos
1.2B-Extract LiquidAI/LFM2-1.2B-Extract temp=0, do_sample=False temp=0 -- -- NO Yes Single-turn, schema prompt
350M-Extract LiquidAI/LFM2-350M-Extract temp=0, do_sample=False temp=0 -- -- NO Yes Single-turn, schema prompt
1.2B-RAG LiquidAI/LFM2-1.2B-RAG temp=0, do_sample=False temp=0 -- -- NO Yes Single-turn
2.6B-Transcript LiquidAI/LFM2-2.6B-Transcript temp=0.3 temp=0.3 -- -- NO Yes Structured input
ColBERT-350M LiquidAI/LFM2-ColBERT-350M PyLate only NO NO NO NO NO Late-interaction retrieval
350M-ENJP-MT LiquidAI/LFM2-350M-ENJP-MT Standard Standard -- -- Yes Yes System prompt controls direction
350M-Math LiquidAI/LFM2-350M-Math Standard Standard -- -- NO Yes
350M-PII-JP LiquidAI/LFM2-350M-PII-Extract-JP temp=0 temp=0 -- -- NO NO Japanese-specific
1.2B-Tool LiquidAI/LFM2-1.2B-Tool -- -- -- -- NO Yes Deprecated, migration guide

Version Pins

Dependency Version Reason
transformers >=5.2.0 VL models require v5; standard models work with v5 too
vLLM ==0.19.0 Latest stable with LFM support (text and vision)
llama.cpp Latest via brew install or b7075+ binaries

Known Limitations

  • Ollama + VL models: Does NOT work. PR pending: ollama/ollama#14069
  • vLLM + VL models: Requires custom precompiled wheel (not standard pip install)
  • Audio models: Use liquid-audio library, not standard Transformers
  • Audio + llama.cpp: Requires special binary (llama-liquid-audio-cli) with mmproj/vocoder files

Update Checklist

When updating a version pin or code pattern:

  1. Update the relevant snippet file(s) in snippets/quickstart/
  2. Update the corresponding main inference page in docs/inference/
  3. Update this matrix if version pins changed
  4. Nano/Audio/ColBERT pages: update inline code manually
  5. Run npx mintlify dev and spot-check a model page