20 lines
579 B
Plaintext
20 lines
579 B
Plaintext
# Dependencies for hf_to_nvfp4_gguf.py, ollama_nvfp4.py,
|
|
# safetensors_to_nvfp4_gguf.py, and nvfp4gguf.
|
|
#
|
|
# llama.cpp (GGUF writer and Hugging Face conversion) and convert_to_quant
|
|
# (LearnedNVFP4Converter) are local checkouts, not pip packages. Point the
|
|
# scripts at them with --llama-cpp and --convert-to-quant.
|
|
#
|
|
# Install a CUDA build of torch that matches your GPU. A CPU wheel works
|
|
# with --device cpu.
|
|
|
|
numpy>=2.2.6
|
|
torch>=2.6.0
|
|
safetensors>=0.4.2
|
|
tqdm>=4.27
|
|
PyYAML>=5.1
|
|
transformers>=4.46
|
|
sentencepiece>=0.1.98,<0.3.0
|
|
protobuf>=4.21.0,<5.0.0
|
|
prodigy-plus-schedule-free
|