Views
No views yet
| Name | Quant method | Size |
|---|---|---|
| layer-skip-v0.Q2_K.gguf | Q2_K | 0.54GB |
| layer-skip-v0.IQ3_XS.gguf | IQ3_XS | 0.58GB |
| layer-skip-v0.IQ3_S.gguf | IQ3_S | 0.6GB |
| layer-skip-v0.Q3_K_S.gguf | Q3_K_S | 0.6GB |
| layer-skip-v0.IQ3_M.gguf | IQ3_M | 0.61GB |
| layer-skip-v0.Q3_K.gguf | Q3_K | 0.64GB |
| layer-skip-v0.Q3_K_M.gguf | Q3_K_M | 0.64GB |
| layer-skip-v0.Q3_K_L.gguf | Q3_K_L | 0.68GB |
| layer-skip-v0.IQ4_XS.gguf | IQ4_XS | 0.7GB |
| layer-skip-v0.Q4_0.gguf | Q4_0 | 0.72GB |
| layer-skip-v0.IQ4_NL.gguf | IQ4_NL | 0.72GB |
| layer-skip-v0.Q4_K_S.gguf | Q4_K_S | 0.72GB |
| layer-skip-v0.Q4_K.gguf | Q4_K | 0.75GB |
| layer-skip-v0.Q4_K_M.gguf | Q4_K_M | 0.75GB |
| layer-skip-v0.Q4_1.gguf | Q4_1 | 0.77GB |
| layer-skip-v0.Q5_0.gguf | Q5_0 | 0.83GB |
| layer-skip-v0.Q5_K_S.gguf | Q5_K_S | 0.83GB |
| layer-skip-v0.Q5_K.gguf | Q5_K | 0.85GB |
| layer-skip-v0.Q5_K_M.gguf | Q5_K_M | 0.85GB |
| layer-skip-v0.Q5_1.gguf | Q5_1 | 0.89GB |
| layer-skip-v0.Q6_K.gguf | Q6_K | 0.95GB |
| layer-skip-v0.Q8_0.gguf | Q8_0 | 1.23GB |
class LayerSkipSFTTrainer(SFTTrainer):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.early_exit_layer = 1
def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=None):
labels = inputs.pop("labels")
outputs = model(**inputs, output_hidden_states=True)
hidden_state = outputs["hidden_states"][self.early_exit_layer]
logits = model.lm_head(hidden_state)
loss = model.loss_function(logits=logits, labels=labels, vocab_size=model.vocab_size)
self.early_exit_layer = (self.early_exit_layer + 1) % model.config.num_hidden_layers
return loss