chore: renormalize cache/metadata.json line endings

fix: make rerank failures diagnosable; drop dead llama-server startup

Merge Space-side changes (style bgcolor, PEP 668 pip flags)

fix: correct answer mangling and silent rerank no-op; drop CSBS branding

fix: add --break-system-packages to pip installs for PEP 668 compliance

matching bgcolor with core site

fix: skip all cache validation, trust embeddings.npy if row count matches

fix: remove source_hash cache check to fix cross-OS cache portability

recomputed indexes

fix: use content hash + relative path in cache validation for portability across deployments

added precomputed index files to accelerate start up

added webhelp-responsive folder

fix: use full path /app/llama-server since /app is not in PATH

fix: clear base image ENTRYPOINT so entrypoint.sh runs as CMD not as argument to llama-server

fix: remove --break-system-packages flag incompatible with pip 22 on Ubuntu 22.04

refactor: use llama.cpp server image as base, install Python on top

fix: correct llama-server binary path to /app/llama-server in multi-stage build

refactor: use ghcr.io/ggml-org/llama.cpp:server image for current llama-server binary

fix: build llama-cpp-python from latest HEAD to support recent unsloth GGUF format

fix: reduce n_ctx to 4096 and n_batch to 256 to fit HF Spaces memory

fix: copy blob to .gguf path instead of symlink to avoid permission issues

fix: symlink blob to .gguf path so llama-cpp-python recognises format

fix: pass resolved blob path to llama-server, bypassing HF symlink

debug: add file size/type/symlink diagnostics before model load

fix: install hf_transfer and enable HF_HUB_ENABLE_HF_TRANSFER for xet GGUF downloads

fix: pre-download model via hf_hub_download to get absolute path, avoid ./ corruption

fix: set HF_HOME to /tmp/hf-cache to avoid /.cache permission error

fix: use --model for gguf filename, not --hf_model_repo_filename

fix: correct model repo ID, filename, and alias to match HF GGUF URL

fix: replace llama-server binary with llama-cpp-python[server] for platform compatibility

fix: write llama-server log and model cache to /tmp, fix /app/cache ownership

fix: correct llama.cpp release asset URL and extraction path

Final review fixes, pre-built cache, HF Spaces deployment config

added source with fix1

initial commit