Model display names are incorrect #49

Activity

1. tomzhu1024

   added

   bugSomething isn't workingSomething isn't working

   on Oct 23, 2026
2. yuseferi commented on Oct 26, 2026

   

   yuseferi

   on Oct 26, 2026

   Owner

   More actions

   Thanks for the report. #50 fixes all four ids listed here:

   • deepseek-v3.2 → DeepSeek V3.2
   • glm-5.3 → GLM 5.3
   • kimi-k2.5 → Kimi K2.5
   • minimax-m2.5 → MiniMax M2.5

   It's waiting on one change. The new letter-prefixed version rule also matches quantization tokens, so mistral-7b-instruct-q4_0 would show as Q4.0 (details in the review on #50). Once that's fixed, merging #50 closes this issue.

   Two related cases that #50 doesn't cover yet, and that you may hit next:

   • deepseek-v2.5-1210 still shows DeepSeek V2 5 1210, because a digits-only token after the pair blocks the join.
   • qwen2.5-72b-instruct still shows Qwen2 5 72B Instruct, because the version is glued to the brand name.

   They can go in #50 or a follow-up.
3. tomzhu1024 commented on Oct 27, 2026

   

   tomzhu1024

   on Oct 27, 2026

   Last edited by tomzhu1024

   Author

   More actions

   Thanks for the review!

   I used an include list rather than an exclude. The pattern now accepts only the real version letters (DeepSeek's v, Kimi's k, MiniMax's m), and keeps quantization suffixes such as q and iq out. The requested mistral-7b-instruct-q4_0 test case is added and the potentially-confusing dall-e-3 test case is removed.

   The minor cases are beyond this PR's scope and behave the same on main, so I've left them for a follow-up.