# KB Framework - HuggingFace Transformers LLM Engine Dependencies
#
# Install: pip install -r requirements-transformers.txt
#
# These dependencies are optional and only needed when using
# TransformersEngine (model_source="huggingface").
# They are NOT required for OllamaEngine.
#
# Core requirements.txt must also be installed.

-r requirements.txt

# ── Core LLM Engine ──────────────────────────────────────────────
# PyTorch for model inference
# Used by: kb/biblio/engine/transformers_engine.py (lazy import)
torch>=2.1

# HuggingFace Transformers for model loading and generation
# Used by: kb/biblio/engine/transformers_engine.py (AutoModelForCausalLM, AutoTokenizer, etc.)
transformers>=4.36

# ── Quantization (optional - Linux/CUDA only) ────────────────────
# 4-bit/8-bit quantization via bitsandbytes
# Used by: kb/biblio/engine/transformers_engine.py (BitsAndBytesConfig)
# Uncomment if you need quantization:
# bitsandbytes>=0.41

# ── Multi-GPU (optional) ─────────────────────────────────────────
# Multi-GPU and device_map="auto" via accelerate
# Used by: kb/biblio/engine/transformers_engine.py (lazy import)
# Uncomment if you need multi-GPU support:
# accelerate>=0.25

# ── Tokenizer backends (model-dependent, optional) ────────────────
# Some models require sentencepiece or protobuf:
# sentencepiece>=0.1
# protobuf>=3.20