deepseek-ai/DeepSeek-V3.1
Full fine-tune vs deepseek-ai/DeepSeek-V3.1-Base. Fine-tune: 100% of the weights changed, across 62 of 62 layers.
Tensors changed: Embeddings 2/2, Attention 551/620, Norms 252/252, MLP 1944/2123, lm_head 1/1, other 2/2
10 derivatives checked
- unsloth/DeepSeek-V3.1-GGUF: Quantization
- bartowski/deepseek-ai_DeepSeek-V3.1-GGUF: Quantization
- nvidia/DeepSeek-V3.1-NVFP4: Different architecture
- QuantTrio/DeepSeek-V3.1-AWQ: Different architecture
- DevQuasar/deepseek-ai.DeepSeek-V3.1-GGUF: Quantization
- mlx-community/DeepSeek-V3.1-4bit: Different architecture
- yujiepan/deepseek-v3.1-tiny-random: Quantization
- mlx-community/DeepSeek-V3.1-8bit: Different architecture
- bullerwins/DeepSeek-V3.1-GGUF: Quantization
- unsloth/DeepSeek-V3.1: Identical copy