Dogge

/

alpaca-13b

like 28

full fine tune with datasets

Thanks, can I ask why the model files are split into so many? And why half are 300mb others 600?

Some weights of the model checkpoint at alpaca-13b were not used when initializing LlamaForCausalLM: ['base_model.model.lm_head.weight']

• This IS expected if you are initializing LlamaForCausalLM from the checkpoint of a model trained on another task or with another architecture (e.g. initializing a BertForSequenceClassification model from a BertForPreTraining model).
• This IS NOT expected if you are initializing LlamaForCausalLM from the checkpoint of a model that you expect to be exactly identical (initializing a BertForSequenceClassification model from a BertForSequenceClassification model).

ValueError: weight is on the meta device, we need a value to put in on 0.

I get this error, do you happen to know how to fix, using this:

model = LlamaForCausalLM.from_pretrained(
"alpaca",
load_in_8bit=True,
torch_dtype=torch.float16,
device_map="auto"
)

I'm also getting this error with just this model, trying to load it in the text-generation-webui in 8bit

env/lib/python3.10/site-packages/accelerate/utils/modeling.py", line 136, in set_module_tensor_to_device
raise ValueError(f"{tensor_name} is on the meta device, we need a value to put in on {device}.")
ValueError: weight is on the meta device, we need a value to put in on 0.

Don't bother, this model is broken beyond repair, also it's not finetuned, it's LoRA.

Don't bother, this model is broken beyond repair, also it's not finetuned, it's LoRA.

Is there a way to check if a model is LoRA, and more in general to get every possible detail out of it, beyond any doubt? I'm very interested in this

The size is a good indication. I'm sure there are other ways, too.