README.md

metadata

language:
  - en
base_model:
  - deepseek-ai/DeepSeek-R1-Distill-Qwen-32B

• Extracted a 64 Rank Lora from DeepSeek-R1-Distill-Qwen-32B
• Merged & Quantized into Q4_K_M

Note: The model seems to be somewhat working with the R1's weird template too but it repeats random Chinese characters and the quality seems to be consistently worse.

Maybe try using the R1 tokenizer.