AutoGuess follow-up: adapter checks

[kallewoof]

ChatML (Phi 4)                  = ChatML (Phi 4)                  : OK      microsoft/phi-4                             
ChatML (Qwen 2.5 based)         = ChatML (Qwen 2.5 based)         : OK      Qwen/Qwen2.5-0.5B-Instruct                  
ChatML (Kimi)                   = ChatML (Kimi)                   : OK      moonshotai/Kimi-K2-Instruct                 
Google Gemma 2                  = Google Gemma 2                  : OK      Efficient-Large-Model/gemma-2-2b-it         
Google Gemma 3                  = Google Gemma 3                  : OK      google/gemma-3-4b-it                        
Google Gemma 3n                 = Google Gemma 3n                 : OK      lmstudio-community/gemma-3n-E4B-it-MLX-bf16 
Llama 3.x                       = Llama 3.x                       : OK      Steelskull/L3.3-Shakudo-70b                 
Llama 4                         = Llama 4                         : OK      nvidia/Llama-4-Scout-17B-16E-Instruct-FP8   
Mistral V7 (with system prompt) = Mistral V7 (with system prompt) : OK      Doctor-Shotgun/MS3.2-24B-Magnum-Diamond     
Mistral V3                      = Mistral V3                      : missing expected fragment [/INST]asst_1</s>: [/INST] asst_1</s>[INST] user_2[/INST] mistralai/Mistral-7B-Instruct-v0.3          
GLM-4                           = GLM-4                           : OK      THUDM/glm-4-9b-chat-hf                      
Phi 3.5                         = Phi 3.5                         : OK      microsoft/Phi-3.5-mini-instruct             
Phi 4 (mini)                    = Phi 4 (mini)                    : system role missing expected fragment <|system|>\nSyS-tEm<|end|>\n: <|system|>SyS-tEm<|end|><|user|>user<|end|><|endoftext|> microsoft/Phi-4-mini-instruct               
Cohere (Aya Expanse 32B based)  = Cohere (Aya Expanse 32B based)  : OK      CohereLabs/aya-expanse-32b                  [default]
DeepSeek V2.5                   = DeepSeek V2.5                   : OK      deepseek-ai/DeepSeek-V2.5                   
Jamba                           = Jamba                           : system role missing expected fragment <|bom|><|system|>SyS-tEm<|eom|>: <|startoftext|><|bom|><|system|> SyS-tEm<|eom|><|bom|><|user|> user<|eom|> ai21labs/Jamba-tiny-dev                     
Dots                            = Dots                            : system role missing expected fragment <|system|>\nSyS-tEm<|endofsystem|>\n: <|system|>SyS-tEm<|endofsystem|><|userprompt|>user<|endofuserprompt|> rednote-hilab/dots.llm1.inst                
RWKV World                      = MISSING                         : ? fla-hub/rwkv7-1.5B-world                    
Mistral (Generic)               = Mistral (Generic)               : missing expected fragment [/INST]\nasst_1</s>: [/INST]asst_1</s>[INST]user_2[/INST] mistralai/Mistral-Nemo-Instruct-2407        
ChatML (Generic)                = ChatML (Generic)                : OK      NewEden/Gemma-27B-chatml                    
There were 6 failure(s)!

Some of these failures are, I believe, actual issues with adapters. I don't know how big of an impact they have, and/or if it's the chat templates that are actually broken, but even then, I think aligning and then fixing is the right course of action.

This builds on top of, and includes, #1650. It should probably not be merged before that one.

[kallewoof]

FWIW, I intentionally put this in here to ensure people didn't just stack more adapters at the end. That said, with these tests, it should error on the right occasions so no big deal.

[LostRuins]

You might want to rebase this off latest.

I am ok with the rwkv keyword change. I am not even sure what the correct one is - half the so called rwkv models are just using chatml so I suspect the template isnt even valid and should be removed.

[kallewoof]

You might want to rebase this off latest.

Done. Also updated kallewoof#1 to run rebased variant.

I am ok with the rwkv keyword change. I am not even sure what the correct one is - half the so called rwkv models are just using chatml so I suspect the template isnt even valid and should be removed.

I think that will work fine, especially if we include e.g. "User:" in the search strings. ChatML content will be matched with the ChatML adapter(s).

[LostRuins]

will merge this first while we reexamine the templates

[LostRuins]

One thing I have learned previously: Not all official templates are ideal either. One major reason we deviated from using Jinja is because often times our own tried-and-tested templates outperform the official ones. Just something to keep in mind.

[kallewoof]

One thing I have learned previously: Not all official templates are ideal either. One major reason we deviated from using Jinja is because often times our own tried-and-tested templates outperform the official ones. Just something to keep in mind.

Absolutely. I think knowing when we deviate is a good idea though. ;)