cognitivecomputations

/

WizardLM-Uncensored-Falcon-7b

like 55

Hi, I have tried to quantize the model to 4 bits without success, following the same procedure I use for the original model tiiuae/falcon-7b. When I checked the model architecture they look quite similar but for some reason the ehartford/WizardLM-Uncensored-Falcon-7b have one extra dimension (I believe is related with one extra token). Do you know why?

As a reference, I leave you the architectures, look at the embedding layer.

tiiuae/falcon-7b:

RWForCausalLM(
  (transformer): RWModel(
    (word_embeddings): Embedding(65024, 4544)
    (h): ModuleList(
      (0-31): 32 x DecoderLayer(
        (input_layernorm): LayerNorm((4544,), eps=1e-05, elementwise_affine=True)
        (self_attention): Attention(
          (maybe_rotary): RotaryEmbedding()
          (query_key_value): Linear4bit(in_features=4544, out_features=4672, bias=False)
          (dense): Linear4bit(in_features=4544, out_features=4544, bias=False)
          (attention_dropout): Dropout(p=0.0, inplace=False)
        )
        (mlp): MLP(
          (dense_h_to_4h): Linear4bit(in_features=4544, out_features=18176, bias=False)
          (act): GELU(approximate='none')
          (dense_4h_to_h): Linear4bit(in_features=18176, out_features=4544, bias=False)
        )
      )
    )
    (ln_f): LayerNorm((4544,), eps=1e-05, elementwise_affine=True)
  )
  (lm_head): Linear(in_features=4544, out_features=65024, bias=False)
)

ehartford/WizardLM-Uncensored-Falcon-7b:

RWForCausalLM(
  (transformer): RWModel(
    (word_embeddings): Embedding(65025, 4544)
    (h): ModuleList(
      (0-31): 32 x DecoderLayer(
        (input_layernorm): LayerNorm((4544,), eps=1e-05, elementwise_affine=True)
        (self_attention): Attention(
          (maybe_rotary): RotaryEmbedding()
          (query_key_value): Linear4bit(in_features=4544, out_features=4672, bias=False)
          (dense): Linear4bit(in_features=4544, out_features=4544, bias=False)
          (attention_dropout): Dropout(p=0.0, inplace=False)
        )
        (mlp): MLP(
          (dense_h_to_4h): Linear4bit(in_features=4544, out_features=18176, bias=False)
          (act): GELU(approximate='none')
          (dense_4h_to_h): Linear4bit(in_features=18176, out_features=4544, bias=False)
        )
      )
    )
    (ln_f): LayerNorm((4544,), eps=1e-05, elementwise_affine=True)
  )
  (lm_head): Linear(in_features=4544, out_features=65025, bias=False)