Error
from mlx_lm import load, generate
model, tokenizer = load("mlx-community/c4ai-command-r7b-12-2024-bf16")
prompt="hello"
if hasattr(tokenizer, "apply_chat_template") and tokenizer.chat_template is not None:
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
response = generate(model, tokenizer, prompt=prompt, verbose=True)
--
says
File /opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/mlx_lm/utils.py:121, in _get_classes(config)
120 try:
--> 121 arch = importlib.import_module(f"mlx_lm.models.{model_type}")
122 except ImportError:
File /opt/anaconda3/envs/f1llama/lib/python3.10/importlib/init.py:126, in import_module(name, package)
125 level += 1
--> 126 return _bootstrap._gcd_import(name[level:], package, level)
File :1050, in _gcd_import(name, package, level)
File :1027, in find_and_load(name, import)
File :1004, in find_and_load_unlocked(name, import)
ModuleNotFoundError: No module named 'mlx_lm.models.cohere2'
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
Cell In[6], line 3
1 from mlx_lm import load, generate
----> 3 model, tokenizer = load("mlx-community/c4ai-command-r7b-12-2024-bf16")
...
124 logging.error(msg)
--> 125 raise ValueError(msg)
127 return arch.Model, arch.ModelArgs
ValueError: Model type cohere2 not supported.
@Rafii
, they added support 2 hours ago. you should wait for the next mlx-lm
release or add it yourself. Check commit here:
Woah, Thanks mate, that was a perfect help. I manually integrated, its processing now.
Also, I was trying to use mlx model to host it on spaces,
this error I have been trying to resolve for 3 hours now,
terminate called after throwing an instance of 'std::runtime_error'
what(): [Matmul::eval_cpu] Currently only supports float32.
--
My model is FP16 U32, that is not supported in spaces, wondering if you came accross this issue. I see you have spaces as well on your profile.
thank you
Yes, but the thing is I already finetuned 8 bit quant model on my data, and I will have to do it again. And finetuning default models on macbook air m3, is a bit challenging, have not done i before.
thanks though
oh, i see.
guess i won't be able to help here, just exploring the model and mlx myself. this repo conversion might be helpful for you though
https://huggingface.co/spaces/mlx-community/mlx-my-repo
Or maybe autotrain mode in model card might be useful too