Closed JinSeoungwoo closed 9 months ago
make sure that generate_prompt(x)
have to return string
make sure that
generate_prompt(x)
have to return string
generate_prompt return string
def generate_prompt(data_point):
full_prompt = prompter.generate_prompt(
data_point["instruction"],
data_point["input"],
data_point["output"],
)
return full_prompt
I think
outputs = self.module.apply(
inputs,
jnp.array(input_ids, dtype="i4"),
jnp.array(attention_mask, dtype="i4"),
jnp.array(position_ids, dtype="i4"),
not train,
None,
False,
output_attentions,
output_hidden_states,
return_dict,
rngs=rng_s,
mutable=mutable,
)
this code from modelling_mistral_flax.py is problem
make sure that
generate_prompt(x)
have to return stringgenerate_prompt return string
def generate_prompt(data_point): full_prompt = prompter.generate_prompt( data_point["instruction"], data_point["input"], data_point["output"], ) return full_prompt
I think
outputs = self.module.apply( inputs, jnp.array(input_ids, dtype="i4"), jnp.array(attention_mask, dtype="i4"), jnp.array(position_ids, dtype="i4"), not train, None, False, output_attentions, output_hidden_states, return_dict, rngs=rng_s, mutable=mutable, )
this code from modelling_mistral_flax.py is problem
removed None and it looks working. but Initializer expected to generate shape (1024, 4096) but got shape (4096, 1024) instead for parameter "kernel" in "/model/layers/remat(0)/self_attn/k_proj"
ill fix that in next commit
Fixed <3 please remove the code you have edited clone the repo or install it with
pip install git+https://github.com/erfanzar/EasyDeL
and transform weights again
Below is the code I used for making train_data
Error :
and also there is a o_proj error in mistral