fix(model_configs): point qwen2.5 OpenVINO models at their real HF repos (#1306)

[Anai-Guo]

What this fixes

Fixes #1306.

load_model("qwen2.5-3b-instruct-ov") fails with:

RepositoryNotFoundError: 401 Client Error.
Repository Not Found for url: https://huggingface.co/api/models/llmware/qwen2-3b-instruct-ov/revision/main.

The requested model name is qwen2.5-…, but the repo in the error is qwen2-…. The catalog entry's hf_repo is missing the .5, so pull_snapshot_from_hf() requests a repo that does not exist. Each affected entry's own link field already points at the correct repo, which is what makes the mismatch unambiguous:

Entry	hf_repo (before)	link / actual repo
qwen2.5-0.5b-instruct-ov (×2)	llmware/qwen2-0.5b-instruct-ov → 401	llmware/qwen2.5-0.5b-instruct-ov → 200
qwen2.5-3b-instruct-ov	llmware/qwen2-3b-instruct-ov → 401	llmware/qwen2.5-3b-instruct-ov → 200

Three lines in global_model_repo_catalog_list, no other change. Verified against the HF API: the repos named before this patch return 401 (do not exist), the ones it points to return 200. All other qwen2.5-*-ov entries in the catalog were already correct.

Note for maintainers (not changed here)

While verifying, four other entries have a link that 404s while their hf_repo is correct — cosmetic, and out of scope for this fix, but worth a follow-up:

• dolphin-2.9.4-llama3.1-8b-ov → link says llmware/dolphi-… (missing n)
• intel-neural-chat-7b-v3-2-ov → link says …-v3-ov
• llama-3.2-1b-instruct-npu-ov → link says …-1b-npu-instruct-ov
• qwen2.5-coder-7b-instruct-ov → link says …-7b-coder-instruct-ov

🤖 Generated with Claude Code