Erreur lors de la conversion des types utilisant la V2 de LLama

[pereconteur]

Lors de la conversion, il semble que le tokenizer et les poids ne correspondent pas : voici mon erreur en utilisant python3.10, vigogne-2-7b-chat :

Loading model file models/vigogne-2-7b-chat_path/consolidated.00.pth
params = Params(n_vocab=-1, n_embd=4096, n_layer=32, n_ctx=2048, n_ff=11008, n_head=32, n_head_kv=32, f_norm_eps=1e-06, rope_scaling_type=None, f_rope_freq_base=None, f_rope_scale=None, n_orig_ctx=None, rope_finetuned=None, ftype=None, path_model=PosixPath('models/vigogne-2-7b-chat_path'))
Loading vocab file 'models/tokenizer.model', type 'spm'
tok_embeddings.weight                            -> token_embd.weight                        | F16    | [32000, 4096]
layers.0.attention.wq.weight                     -> blk.0.attn_q.weight                      | F16    | [4096, 4096]
layers.0.attention.wk.weight                     -> blk.0.attn_k.weight                      | F16    | [4096, 4096]
[...]
output.weight                                    -> output.weight                            | F16    | [32000, 4096]
Writing models/vigogne-2-7b-chat_path/ggml-model-f16.gguf, format 1
Traceback (most recent call last):
  File "/Users/pereconteur/Desktop/IA/vigogne/scripts/../../llama.cpp/convert.py", line 1204, in <module>
    main()
  File "/Users/pereconteur/Desktop/IA/vigogne/scripts/../../llama.cpp/convert.py", line 1199, in main
    OutputFile.write_all(outfile, ftype, params, model, vocab, special_vocab, concurrency = args.concurrency, endianess=endianess)
  File "/Users/pereconteur/Desktop/IA/vigogne/scripts/../../llama.cpp/convert.py", line 909, in write_all
    check_vocab_size(params, vocab)
  File "/Users/pereconteur/Desktop/IA/vigogne/scripts/../../llama.cpp/convert.py", line 796, in check_vocab_size
    raise Exception(msg)
Exception: Vocab size mismatch (model has -1, but models/tokenizer.model has 32000).
Traceback (most recent call last):
  File "/Users/pereconteur/Desktop/IA/Vigogne_auto_installer/vigogne_V2_auto_install_Mac_Linux.py", line 269, in <module>
    check_requirements()
  File "/Users/pereconteur/Desktop/IA/Vigogne_auto_installer/vigogne_V2_auto_install_Mac_Linux.py", line 263, in check_requirements
    goForVigogneInstallation(model_vig, poids_B, type_vig)
  File "/Users/pereconteur/Desktop/IA/Vigogne_auto_installer/vigogne_V2_auto_install_Mac_Linux.py", line 205, in goForVigogneInstallation
    subprocess.run(commandConvert, check=True)
  File "/Users/pereconteur/miniconda3/lib/python3.10/subprocess.py", line 526, in run
    raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command '['python3.10', '../../llama.cpp/convert.py', 'models/vigogne-2-7b-chat_path']' returned non-zero exit status 1.