ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update mistral.py

Browse source
This commit is contained in:
Daniel Han-Chen 2024-02-03 19:30:22 +11:00
commit d76f583349
1 changed files with 21 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

42
unsloth/models/mistral.py
View file

@ -201,31 +201,31 @@ def MistralForCausalLM_fast_forward(
# decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
self.model._has_no_labels = labels is None
if past_key_values is not None and \
hasattr(self.model.layers[0].self_attn, "paged_attention"):
outputs = LlamaModel_fast_forward_inference(
self.model,
input_ids,
past_key_values,
)
else:
outputs = self.model(
input_ids=input_ids,
causal_mask=causal_mask,
attention_mask=attention_mask,
position_ids=position_ids,
past_key_values=past_key_values,
inputs_embeds=inputs_embeds,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
# if past_key_values is not None and \
# hasattr(self.model.layers[0].self_attn, "paged_attention"):
# outputs = LlamaModel_fast_forward_inference(
# self.model,
# input_ids,
# past_key_values,
# )
# else:
outputs = self.model(
input_ids=input_ids,
causal_mask=causal_mask,
attention_mask=attention_mask,
position_ids=position_ids,
past_key_values=past_key_values,
inputs_embeds=inputs_embeds,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states,
return_dict=return_dict,
)
pass
hidden_states = outputs[0]
bsz, q_len, hd = hidden_states.shape
if bsz == 1 and q_len == 1:
if False:#bsz == 1 and q_len == 1:
logits = torch.mv(self.lm_head.weight, hidden_states.ravel())
logits = logits.unsqueeze(0).unsqueeze(0)
else: