[Bug] mlc-ai-nightly-cpu==0.26.dev203 cannot install due to apache-tvm-ffi>=0.1.13 version conflict on PyPI

Bug Report

pip install mlc-ai-nightly-cpu==0.26.dev203 fails because it requires apache-tvm-ffi>=0.1.13, but the latest version on PyPI is apache-tvm-ffi==0.1.13rc1 (a pre-release). Under PEP 440, pre-release versions do not satisfy a bare >=0.1.13 specifier unless the specifier itself includes a pre-release marker. This makes the installation impossible without --no-deps.

To Reproduce

conda create -n mlc python=3.13 -y
conda activate mlc
python -m pip install --pre -U -f https://mlc.ai/wheels mlc-ai-nightly-cpu

Output

ERROR: Could not find a version that satisfies the requirement apache-tvm-ffi>=0.1.13 (from mlc-ai-nightly-cpu) (from versions: ..., 0.1.13rc1)
ERROR: No matching distribution found for apache-tvm-ffi>=0.1.13

Workaround

pip install --pre --no-deps -U -f https://mlc.ai/wheels mlc-ai-nightly-cpu==0.26.dev203
pip install apache-tvm-ffi==0.1.13rc1

Expected Behavior

pip install --pre -f https://mlc.ai/wheels mlc-ai-nightly-cpu

should succeed without needing --no-deps.

Root Cause

apache-tvm-ffi==0.1.13rc1 on PyPI is a PEP 440 pre-release (rc1). The dependency in mlc-ai-nightly-cpu==0.26.dev203 is:

apache-tvm-ffi>=0.1.13

Per PEP 440, 0.1.13rc1 does not satisfy >=0.1.13 because pip requires either:

• A stable release >=0.1.13 (e.g., 0.1.13, 0.1.14), or
• The specifier to explicitly allow pre-releases (e.g., >=0.1.13rc1)

Since no stable 0.1.13 exists yet, pip cannot resolve the dependency.

Suggested Fix

Either:

1. Publish a stable apache-tvm-ffi==0.1.13 release
2. Change the dependency specifier to apache-tvm-ffi>=0.1.13rc1
3. Host apache-tvm-ffi wheels on mlc.ai/wheels directly

Environment

• OS: macOS 15.7.4 (Apple Silicon M4)
• Python: 3.13 (Miniconda)
• Package versions: mlc-ai-nightly-cpu==0.26.dev203, apache-tvm-ffi==0.1.13rc1