No longer works with current version of llama.cpp

#1
by dzupin - opened

Hi
In my testing this was the best Llame3 full fine-tune in GGUF format. But after update of llama.cpp to better support Lllama3 tokenizer (about a week ago) output produced by this model is broken (garbled or empty)
Did anybody else tried to run it with current version of llama.cpp ? Any idea if it can be fixed ?

In log I can see following:

llama_model_loader: - type f32: 65 tensors
llama_model_loader: - type q6_K: 226 tensors
llm_load_vocab: missing pre-tokenizer type, using: 'default'
llm_load_vocab:
llm_load_vocab: ************************************
llm_load_vocab: GENERATION QUALITY WILL BE DEGRADED!
llm_load_vocab: CONSIDER REGENERATING THE MODEL
llm_load_vocab: ************************************
llm_load_vocab:
llm_load_vocab: mismatch in special tokens definition ( 259/128260 vs 260/128260 ).
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = llama

Example of output produced by model:
****** Retrieved OUTPUT:
·onononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononononlieronononononononononononononononononlierlierlierlierlieronlierlierlierlylierlyersonononononononononononononononononononononononononlierHSlieronHSlieronlierlierlieronlierly
<|end_of_text|>ononononononononlierlierlylieronlieronlierlieronHSlieronlieronlierlieronlieronononononononononononononononononononononononononononononononononononononononlierlierlieronlierlier️lierlierlierlylierHSlier
ersonononononononononononononononononononononononononononlierlieronlierlierlierlier
awaiawaiawaionononononlierlylierlieronODEvs<|end_of_text|>onononononlierlierlier️ for due
lierly
awaiawaiawaionononononononononononononononononononononononononononononononononononononononononononlierlier
ersonawaionEdgelierHSlierHSlierlierely️ due
due
ode
lierlyarchononawaiawaionononawaiononawaiawaiawaiawaiersonon
due<|end_of_text|>ononononononononononononononononlierlierlierly
onlierlierHS
uses its dueawaiawaiawaiawaiawaiawaion
vs️️ononawaiawaiawaiawaiononawaiononawaiawaiawaiawaionon.githubonawaiawaiawaiawaiawaiCompat<|end_of_text|>ODEvsCompaterson
lierly
awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai
.github due due<|end_of_text|>
onCompatannyadue
onlierly dueawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiCompaton️.github
awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai.githubononODEvsdue<|end_of_text|>.githubonawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai
awaiawaiawaiawaiawaiCompatawai
.github due due
awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai.github

onarchawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiCompat<|end_of_text|>ODEHS
awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai.githubCompat<|end_of_text|>onon
awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai.githubawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|>onawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai.github<|end_of_text|>
awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|>
awaiawai.githubawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|>awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|>awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|>awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|><|end_of_text|> dueawaiawaiawai<|end_of_text|>awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|>awaiawaiawai<|end_of_text|>awaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawaiawai<|end_of_text|><|im_end|>

Ah yes this is an older quant, but worth revisiting, i'll make an updated one today :)

Wow that was fast!
Thank you very much for updating the model. I really like this one.

Sign up or log in to comment