biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CORE] Quantized lm-head Framework (#4442)

Browse Source 
Co-authored-by: Robert Shaw <rshaw@neuralmagic.com>
Co-authored-by: ZX <zx@lbx.dev>
This commit is contained in:
Qubitium-ModelCloud
2024-07-03 06:25:17 +08:00
committed by GitHub
parent 7c008c51a9
commit ee93f4f92a
48 changed files with 268 additions and 121 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/models/phi.py
View File

@@ -268,7 +268,8 @@ class PhiForCausalLM(nn.Module, SupportsLoRA):
self.lm_head = ParallelLMHead(config.vocab_size,
config.hidden_size,
bias=True)
bias=True,
quant_config=quant_config)
self.logits_processor = LogitsProcessor(config.vocab_size)
self.sampler = Sampler()
@@ -287,7 +288,7 @@ class PhiForCausalLM(nn.Module, SupportsLoRA):
def compute_logits(self, hidden_states: torch.Tensor,
sampling_metadata: SamplingMetadata) -> torch.Tensor:
logits = self.logits_processor(self.lm_head.weight, hidden_states,
logits = self.logits_processor(self.lm_head, hidden_states,
sampling_metadata, self.lm_head.bias)
return logits