LiquidAI/LFM2.5-VL-3B
Full fine-tune (fewer layers) vs LiquidAI/LFM2.5-2.6B-Base. LiquidAI/LFM2.5-2.6B-Base matches part of LiquidAI/LFM2.5-VL-3B (266 of its 707 tensors; the rest of LiquidAI/LFM2.5-VL-3B has no counterpart). On the shared part: Fine-tune: 100% of the weights changed, across 30 of 30 layers.
Tensors changed: Embeddings 2/2, other 66/66, MLP 90/90, Norms 76/76, Attention 32/32
24 derivatives checked
- unsloth/LFM2.5-VL-3B-GGUF: Same weights, converted (fewer layers)
- LiquidAI/LFM2.5-VL-3B-GGUF: Same weights, converted (fewer layers)
- LiquidAI/d1-3B: Partial fine-tune
- SC117/LFM2.5-VL-3B-Uncensored-GGUF: Abliterated (fewer layers)
- LiquidAI/LFM2.5-VL-3B-MLX-4bit: Quantization
- mlx-community/LFM2.5-VL-3B-OptiQ-4bit: Quantization
- bartowski/LiquidAI_LFM2.5-VL-3B-GGUF: Same weights, converted (fewer layers)
- unsloth/LFM2.5-VL-3B: Identical copy
- useful-quants/LFM2.5-VL-3B-NVFP4-FP8-Mixed: Quantization
- NANI-Nithin/LFM2.5-VL-3B-GGUF: Same weights, converted (fewer layers)
- LiquidAI/LFM2.5-VL-3B-DSpark: Different architecture
- DJLougen/LFM2.5-VL-3B-DFlash-drafter: Different architecture
- prithivMLmods/LFM2.5-VL-3B-GGUF: Same weights, converted (fewer layers)
- heretic-org/LFM2.5-VL-3B-heretic: Abliterated
- LiquidAI/LFM2.5-VL-3B-MLX-8bit: Quantization
- LiquidAI/LFM2.5-VL-3B-MLX-bf16: Different architecture
- npario/LFM2.5-VL-3B-GGUF: Same weights, converted (fewer layers)
- RKNNAI/RK182X-VLM-LFM2.5-VL-3B: Quantization
- LiquidAI/LFM2.5-VL-3B-MLX-6bit: Quantization
- LiquidAI/LFM2.5-VL-3B-MLX-5bit: Quantization
- fwizzer1/fwizzer-v3-ru: Full fine-tune (fewer layers)
- win10/Lfm2-MaleCNS-Titans-VL-3B: Different architecture
- EryriLabs/LFM2.5-VL-3B-DragOn: Full fine-tune
- MuXodious/LFM2.5-VL-3B-absolute-heresy: Abliterated
2 other repos hold byte-for-byte identical weight files.