Outdated pinned dependencies in tools/ray_tpu/ (TF 2.9.3, JAX 0.3.4, numpy 1.22)

[kredd2506]

Description

The dependency files in tools/ray_tpu/ have pinned versions that are 3+ years old and may have known security vulnerabilities or compatibility issues with modern Python environments.

tools/ray_tpu/src/tune/requirements.txt

Package	Pinned Version	Latest Stable	Age
tensorflow-cpu	2.9.3	2.18+	~3 years
jax	0.3.4	0.4.x+	~4 years
jaxlib	0.3.2	0.4.x+	~4 years
numpy	1.22.0	1.26+	~4 years
protobuf	3.19.0	4.x+	~4 years
flax	0.4.1	0.8+	~3 years
tensorflow-datasets	4.4.0	4.9+	~3 years

tools/ray_tpu/src/serve/requirements.txt

• ray[serve] is pinned to 2.5.1 (June 2023)
• fastapi and pillow have no version pins at all (inconsistent with the tune requirements)

Impact

• Users following the Ray TPU examples may encounter installation failures or incompatibilities with current Python 3.10+ environments
• Older versions of protobuf and numpy have known CVEs
• jax 0.3.x is incompatible with current TPU runtimes

Suggested Fix

Update all pinned versions to current stable releases, or at minimum specify a compatible range (e.g., numpy>=1.22,<2.0).