Closed doyeonyeah closed 2 years ago
Oh sorry about it. I'm updating the model. You can either install from an older commit (like a month old) or wait maybe a week for the new model.
@borisdayma Thanks, it works perfectly with the commit 31b52fa675175a828e4a5d180e8016fa630d67a from 21/11/17.
I'm looking forward to the newest update. Love your work & please keep up the great work!
This has now been fixed. Note: we should have a new inference notebook soon that is much faster.
When running the inference pipeline notebook on Google Collab (free version) an error is raised when loading DALLE-mini model:
# Load dalle-mini
model = DalleBart.from_pretrained(
DALLE_MODEL, revision=DALLE_COMMIT_ID, dtype=dtype, abstract_init=True
)
Error :
wandb: Downloading large artifact model-3e2l7fxk:latest, 1625.33MB. 7 files... Done. 0:0:0
---------------------------------------------------------------------------
AssertionError Traceback (most recent call last)
[/usr/local/lib/python3.7/dist-packages/dalle_mini/model/configuration.py](https://localhost:8080/#) in __init__(self, normalize_text, encoder_vocab_size, image_vocab_size, image_length, max_text_length, encoder_layers, encoder_ffn_dim, encoder_attention_heads, decoder_layers, decoder_ffn_dim, decoder_attention_heads, activation_function, d_model, dropout, attention_dropout, activation_dropout, init_std, scale_embedding, gradient_checkpointing, use_cache, is_encoder_decoder, forced_eos_token_id, tie_word_embeddings, do_sample, use_bias, ln_type, ln_positions, use_head_scale, use_cosine_attention, tau_init, use_deepnet_scaling, use_glu, use_alibi, sinkhorn_iters, use_final_ln_encoder, use_final_ln_decoder, force_ln_scale, **kwargs)
106 assert (
107 use_final_ln_encoder
--> 108 ), "use_final_ln_encoder must be True when ln_positions is 'postln'"
109 assert (
110 use_final_ln_decoder
AssertionError: use_final_ln_encoder must be True when ln_positions is 'postln'
Thanks for the feedback. It should now work.
This is the full output when I ran the code for loading model, after installing all of the prerequisites (transformer version=0.3.5) on Colab.
AttributeError: 'BartConfig' object has no attribute 'image_vocab_size' - I wasn't able to find any examples of this error on the net.
UnfilteredStackTrace Traceback (most recent call last)