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

[QUANT] Add GPTQModel Dynamic Quantization + lm_head Quantization (#3790)

Browse Source 
Signed-off-by: ZX-ModelCloud <zx@modelcloud.ai>
Co-authored-by: ZX-ModelCloud <zx@modelcloud.ai>
This commit is contained in:
Qubitium-ModelCloud
2025-03-05 17:11:00 +08:00
committed by GitHub
parent 583d6af71b
commit 56a724eba3
56 changed files with 1988 additions and 282 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
python/sglang/srt/models/llama_reward.py
View File

@@ -22,6 +22,7 @@ from sglang.srt.layers.pooler import EmbeddingPoolerOutput, Pooler, PoolingType
from sglang.srt.layers.quantization.base_config import QuantizationConfig
from sglang.srt.model_executor.forward_batch_info import ForwardBatch
from sglang.srt.models.llama import LlamaForCausalLM, LlamaModel
from sglang.srt.utils import add_prefix
class LlamaForSequenceClassification(nn.Module):
@@ -29,12 +30,15 @@ class LlamaForSequenceClassification(nn.Module):
self,
config: LlamaConfig,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
) -> None:
super().__init__()
self.config = config
self.quant_config = quant_config
self.num_labels = config.num_labels
self.model = LlamaModel(config, quant_config=quant_config)
self.model = LlamaModel(
config, quant_config=quant_config, prefix=add_prefix("model", prefix)
)
self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
self.pooler = Pooler(pooling_type=PoolingType.LAST, normalize=False)
@@ -82,8 +86,9 @@ class LlamaForSequenceClassificationWithNormal_Weights(LlamaForSequenceClassific
self,
config: LlamaConfig,
quant_config: Optional[QuantizationConfig] = None,
prefix: str = "",
) -> None:
super().__init__(config, quant_config)
super().__init__(config, quant_config, prefix=prefix)
self.weights = self.Weights(config.hidden_size, self.num_labels)
@torch.no_grad()