EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

[FEAT] Support GGUF format (#2215)

Browse Source 
Co-authored-by: Yang Zheng(SW)(Alex) <you@example.com>
This commit is contained in:
Yang Zheng
2024-11-30 16:44:48 +08:00
committed by GitHub
parent 0d6a49bd7d
commit 883c955489
39 changed files with 180 additions and 89 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/sglang/srt/models/minicpm3.py
View File

@@ -585,12 +585,10 @@ class MiniCPM3ForCausalLM(nn.Module):
hidden_states = self.model(input_ids, positions, forward_batch, input_embeds)
hidden_states = hidden_states / self.scale_width
if self.config.tie_word_embeddings:
lm_head_weight = self.model.embed_tokens.weight
lm_head = self.model.embed_tokens
else:
lm_head_weight = self.lm_head.weight
return self.logits_processor(
input_ids, hidden_states, lm_head_weight, forward_batch
)
lm_head = self.lm_head
return self.logits_processor(input_ids, hidden_states, lm_head, forward_batch)
def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
stacked_params_mapping = [