pretrain errors

[linchen111]

When I run pretrain scripts, I got this: File "/data/lc/Multi-image/multi_token/multi_token/language_models/mistral.py", line 85, in forward ) = self.prepare_inputs_labels_for_multimodal( File "/data/lc/Multi-image/multi_token/multi_token/language_models/base_model.py", line 69, in prepare_inputs_labels_for_multimodal m_vals = m.forward(kwargs.get(m.name)) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(*args, kwargs) File "/data/lc/Multi-image/multi_token/multi_token/modalities/vision_clip.py", line 177, in forward image_features.append(self.module.forward(image_batch)) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(*args, *kwargs) File "/data/lc/Multi-image/multi_token/multi_token/modalities/vision_clip.py", line 40, in forward image_forward_outs = self.image_model( File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl return self._call_impl(args, kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl return forward_call(*args, kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/transformers/models/clip/modeling_clip.py", line 925, in forward return self.vision_model( File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl return self._call_impl(*args, *kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl return forward_call(args, kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/transformers/models/clip/modeling_clip.py", line 849, in forward hidden_states = self.embeddings(pixel_values) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl return self._call_impl(*args, kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl return forward_call(*args, kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/transformers/models/clip/modeling_clip.py", line 190, in forward patch_embeds = self.patch_embedding(pixel_values.to(dtype=target_dtype)) # shape = [, width, grid, grid] File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1511, in _wrapped_call_impl return self._call_impl(args, kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1520, in _call_impl return forward_call(*args, kwargs) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/conv.py", line 460, in forward return self._conv_forward(input, self.weight, self.bias) File "/root/miniconda3/envs/multi-image/lib/python3.10/site-packages/torch/nn/modules/conv.py", line 456, in _conv_forward return F.conv2d(input, weight, bias, self.stride, RuntimeError: weight should have at least three dimension

[sshh12]

What's the command and dataset you are using?

[linchen111]

What's the command and dataset you are using?

CUDA_VISIBLE_DEVICES=0 deepspeed scripts/train_model.py \ --model_name_or_path /data/lc/Multi-image/Model/Mistral-7B-Instruct-v0.1 \ --model_cls MistralLMMForCausalLM \ --modality_builder vision_clip \ --dataset_path /data/lc/Multi-image/data/llava_pretrain_data/ok_data \ （this one I use prepare-data.py） --output_dir /data/lc/Multi-image/Model/my_lmm_pretrain \ --pretrain_projectors \ --lora_enable True \ --num_train_epochs 1 \ --gradient_checkpointing True \ --per_device_train_batch_size 1 \ --per_device_eval_batch_size 1 \ --gradient_accumulation_steps 32 \ --model_max_length 1024 \ --evaluation_strategy "no" \ --save_strategy "steps" \ --save_steps 2048 \ --save_total_limit 1 \ --learning_rate 1e-5 \ --weight_decay 0. \ --warmup_ratio 0.03 \ --lr_scheduler_type "cosine" \ --dataloader_num_workers 2 \ --logging_steps 1 \ --deepspeed ./configs/zero3_offload.json

[spydaz]

please create a jupiter notebook for training on google colab please. I was able to merge the multi lora into my mistral model. so essentially disregarding the lora but how to train the model to enforce the data through the new merged tensor model ... i still have the components separate so i can reload from base and lora ... to merge iwth my own model i ha to clone the lora change the basemodel value (mine is a mistral so it is the same tensor array. it merged unsloth with no issues . I just need a jupiter note book to train on unsloth , with the dataset on hf... for the multi, input. i found that when inputting a image on lmstudio it sent 5 or 6 inputs , i guess it was looking for either input , image, doc, sound, video ? i tried it on kobolo it still worked fine as a llm , and when i went to enter the image in the background it was using llava to encode the image , but the context was to small ... so i think a image of only 512,512 might work... ?

its only the traiing methods

[sshh12]

@linchen111 im curious if you are able to run the CLIP demo code with one of your example images

https://huggingface.co/docs/transformers/model_doc/clip

from PIL import Image
import requests

from transformers import CLIPProcessor, CLIPModel

model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")

url = "http://images.cocodataset.org/val2017/000000039769.jpg"
image = Image.open(requests.get(url, stream=True).raw)

inputs = processor(text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True)

outputs = model(**inputs)
logits_per_image = outputs.logits_per_image  # this is the image-text similarity score
probs = logits_per_image.softmax(dim=1)  # we can take the softmax to get the label probabilities