| license: mit | |
| base_model: | |
| - deepseek-ai/DeepSeek-R1 | |
| base_model_relation: quantized | |
| # DeepSeek-R1-W4AFP8 | |
| This model is a mixed-precision quantized DeepSeek-R1, with dense layer using `FP8_BLOCK_SCALING`, MoE layers uses INT4 weights and FP8 activation. |