Mistral-7B-LoRA-VisionCLIP-LLAVA

These are weights for a version of mistralai/Mistral-7B-Instruct-v0.1 finetuned for multimodal applications.

Modalities

• CLIPVisionModality (use <image> in text and provide images, encoded as 576 tokens)

Dataset

/data/llava-finetune-full (544610 examples)

{'id': '000000033471', 'images': ['/data/llava_finetune_data/images/coco/train2017/train2017/000000033471.jpg'], 'messages': [{'content': '<image>\nWhat are the colors of the bus in the image?', 'role': 'user'}, {'content': 'The bus in the image is white and red.', 'role': 'assistant'}, {'content': 'What feature can be seen on the back of the bus?', 'role': 'user'}, {'content': 'The back of the bus features an advertisement.', 'role': 'assistant'}, {'content': 'Is the bus driving down the street or pulled off to the side?', 'role': 'user'}, {'content': 'The bus is driving down the street, which is crowded with people and other vehicles.', 'role': 'assistant'}]}

Training Device(s)

name, pci.bus_id, vbios_version
NVIDIA GeForce RTX 3090 Ti, 00000000:02:00.0, 94.02.a0.00.41

Usage

GitHub: https://github.com/sshh12/multi_token

Model

MistralLMMForCausalLM.model =

PeftModelForCausalLM(
  (base_model): LoraModel(
    (model): MistralLMMForCausalLM(
      (model): MistralLMMModel(
        (embed_tokens): Embedding(32000, 4096)
        (layers): ModuleList(
          (0-31): 32 x MistralDecoderLayer(
            (self_attn): MistralAttention(
              (q_proj): Linear(
                in_features=4096, out_features=4096, bias=False
                (lora_dropout): ModuleDict(
                  (default): Dropout(p=0.05, inplace=False)
                )
                (lora_A): ModuleDict(
                  (default): Linear(in_features=4096, out_features=64, bias=False)
                )
                (lora_B): ModuleDict(
                  (default): Linear(in_features=64, out_features=4096, bias=False)
                )
                (lora_embedding_A): ParameterDict()
                (lora_embedding_B): ParameterDict()
              )
              (k_proj): Linear(
                in_features=4096, out_features=1024, bias=False
                (lora_dropout): ModuleDict(
                  (default): Dropout(p=0.05, inplace=False)
                )
                (lora_A): ModuleDict(
                  (default): Linear(in_features=4096, out_features=64, bias=False)
                )
                (lora_B): ModuleDict(
                  (default): Linear(in_features=64, out_features=1024, bias=False)
                )
                (lora_embedding_A): ParameterDict()
                (lora_embedding_B): ParameterDict()
              )
              (v_proj): Linear(
                in_features=4096, out_features=1024, bias=False
                (lora_dropout): ModuleDict(
                  (default): Dropout(p=0.05, inplace=False)
                )
                (lora_A): ModuleDict(
                  (default): Linear(in_features=4096, out_features=64, bias=False)
                )
                (lora_B): ModuleDict(
                  (default): Linear(in_features=64, out_features=1024, bias=False)
                )
                (lora_embedding_A): ParameterDict()
                (lora_embedding_B): ParameterDict()
              )
              (o_proj): Linear(
                in_features=4096, out_features=4096, bias=False
                (lora_dropout): ModuleDict(
                  (default): Dropout(p=0.05, inplace=False)
                )
                (lora_A): ModuleDict(
                  (default): Linear(in_features=4096, out_features=64, bias=False)
                )
                (lora_B): ModuleDict(
                  (default): Linear(in_features=64, out_features=4096, bias=False)
                )
                (lora_embedding_A): ParameterDict()
                (lora_embedding_B): ParameterDict()
              )
              (rotary_emb): MistralRotaryEmbedding()
            )
            (mlp): MistralMLP(
              (gate_proj): Linear(
                in_features=4096, out_features=14336, bias=False
                (lora_dropout): ModuleDict(
                  (default): Dropout(p=0.05, inplace=False)
                )
                (lora_A): ModuleDict(
                  (default): Linear(in_features=4096, out_features=64, bias=False)
                )
                (lora_B): ModuleDict(
                  (default): Linear(in_features=64, out_features=14336, bias=False)
                )
                (lora_embedding_A): ParameterDict()
                (lora_embedding_B): ParameterDict()
              )
              (up_proj): Linear(
                in_features=4096, out_features=14336, bias=False
                (lora_dropout): ModuleDict(
                  (default): Dropout(p=0.05, inplace=False)
                )
                (lora_A): ModuleDict(
                  (default): Linear(in_features=4096, out_features=64, bias=False)
                )
                (lora_B): ModuleDict(
                  (default): Linear(in_features=64, out_features=14336, bias=False)
                )
                (lora_embedding_A): ParameterDict()
                (lora_embedding_B): ParameterDict()
              )
              (down_proj): Linear(
                in_features=14336, out_features=4096, bias=False
                (lora_dropout): ModuleDict(
                  (default): Dropout(p=0.05, inplace=False)
                )
                (lora_A): ModuleDict(
                  (default): Linear(in_features=14336, out_features=64, bias=False)
                )
                (lora_B): ModuleDict(
                  (default): Linear(in_features=64, out_features=4096, bias=False)
                )
                (lora_embedding_A): ParameterDict()
                (lora_embedding_B): ParameterDict()
              )
              (act_fn): SiLUActivation()
            )
            (input_layernorm): MistralRMSNorm()
            (post_attention_layernorm): MistralRMSNorm()
          )
        )
        (norm): MistralRMSNorm()
        (vision_clip_lmm_projector): Sequential(
          (0): Linear(in_features=1024, out_features=4096, bias=True)
          (1): GELU(approximate='none')
          (2): Linear(in_features=4096, out_features=4096, bias=True)
        )
      )
      (lm_head): Linear(in_features=4096, out_features=32000, bias=False)
    )
  )
)