DeepSeek-V4-Pro-Qwen3.5-9B MTP BF16 archive
Byte-identical archival mirror of publicly released Jackrong model files. No additional fine-tuning or quantization was performed for this mirror, and no private training data or experiment results are included.
- Native Transformers Safetensors shards, tokenizer and configuration:
experiment/native-source/. - BF16 GGUF, with MTP tensors:
experiment/gguf/DeepSeek-V4-Pro-Qwen3.5-9B-MTP-BF16.gguf(18,407,322,720 bytes, approximately 17.1 GiB). This is BF16, not a four-bit quantization. - Full per-file SHA-256 and upstream pinned revisions:
archive-manifest.json.
Native source: https://huggingface.co/Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B/tree/40c1fbe115db8a76ea0c594fc6759d39ffab4d15
GGUF source: https://huggingface.co/Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B-MTP-GGUF/tree/99b43f206bfe6e4e4c311d8fe5bf9b3a9d06d08e
GGUF SHA-256: 995cbe2bb9e0fda577bac0d5f6f0918722663969ab606f7365f15acd7d70db3f.
The upstream model card and license information are preserved in experiment/native-source/README.md. Read that card for model capabilities and limitations. This mirror makes no additional quality claims.