moonshotai/Kimi-VL-A3B-Instruct
Full fine-tune vs moonshotai/Moonlight-16B-A3B. Fine-tune: 100% of the weights changed, across 27 of 27 layers.
Tensors changed: Embeddings 1/1, Norms 82/82, MLP 2446/2446, Attention 135/135, lm_head 1/1
8 derivatives checked
- moonshotai/Kimi-VL-A3B-Thinking: Full fine-tune
- moonshotai/Kimi-VL-A3B-Thinking-2506: Full fine-tune
- mradermacher/Kimi-VL-A3B-Instruct-GGUF: Quantization
- mradermacher/Kimi-VL-A3B-Instruct-i1-GGUF: Quantization
- ssweens/Kimi-VL-A3B-Instruct-GGUF: Same weights, converted
- internetoftim/Kimi-VL-A3B-Instruct-mlx-4bit: Quantization
- Userb1az/Kimi-VL-A3B-Thinking-2506-GGUF: Full fine-tune
- mlx-community/Kimi-VL-A3B-Thinking-2506-bf16: Different architecture
3 other repos hold byte-for-byte identical weight files.