@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 4ea1f9678d
Fork 0

A high-throughput and memory-efficient inference and serving engine for LLMs

Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git

Synced Nov 17, 01:45 UTC

[BugFix] Resolved Issues For LinearMethod --> QuantConfig (#4418)

Robert Shaw committed Apr 27, 2024 at 14:35 UTC 4ea1f9678dd93f02424ab3de2149f83a490e6c6f
10 files changed -10
vllm/model_executor/models/bloom.py
-1
@@ -139,7 +139,6 @@ class BloomMLP(nn.Module):
139 4 * hidden_size,
140 quant_config=quant_config,
141 )
142 - quant_config = getattr(quant_config, "quant_config", None)
142 self.gelu_impl = get_act_fn("gelu", quant_config, 4 * hidden_size)
143 self.dense_4h_to_h = RowParallelLinear(
144 4 * hidden_size,
vllm/model_executor/models/falcon.py
-1
@@ -203,7 +203,6 @@ class FalconMLP(nn.Module):
203 bias=config.bias,
204 skip_bias_add=True,
205 quant_config=quant_config)
206 - quant_config = getattr(quant_config, "quant_config", None)
206 self.act = get_act_fn("gelu", quant_config, 4 * hidden_size)
207 self.reduce_row_parallel_results = not (config.new_decoder_architecture
208 or config.parallel_attn)
vllm/model_executor/models/gpt2.py
-1
@@ -107,7 +107,6 @@ class GPT2MLP(nn.Module):
107 bias=True,
108 quant_config=quant_config,
109 )
110 - quant_config = getattr(quant_config, "quant_config", None)
110 self.act = get_act_fn(config.activation_function, quant_config,
111 intermediate_size)
112
vllm/model_executor/models/gpt_bigcode.py
-1
@@ -128,7 +128,6 @@ class GPTBigMLP(nn.Module):
128 bias=True,
129 quant_config=quant_config,
130 )
131 - quant_config = getattr(quant_config, "quant_config", None)
131 self.act = get_act_fn(config.activation_function, quant_config,
132 intermediate_size)
133
vllm/model_executor/models/gpt_j.py
-1
@@ -120,7 +120,6 @@ class GPTJMLP(nn.Module):
120 hidden_size,
121 quant_config=quant_config,
122 )
123 - quant_config = getattr(quant_config, "quant_config", None)
123 self.act = get_act_fn(config.activation_function, quant_config,
124 intermediate_size)
125
vllm/model_executor/models/gpt_neox.py
-1
@@ -119,7 +119,6 @@ class GPTNeoXMLP(nn.Module):
119 config.hidden_size,
120 quant_config=quant_config,
121 )
122 - quant_config = getattr(quant_config, "quant_config", None)
122 self.act = get_act_fn(config.hidden_act, quant_config,
123 config.intermediate_size)
124
vllm/model_executor/models/mpt.py
-1
@@ -146,7 +146,6 @@ class MPTMLP(nn.Module):
146 bias=not config.no_bias,
147 quant_config=quant_config,
148 )
149 - quant_config = getattr(quant_config, "quant_config", None)
149 self.act = get_act_fn("gelu", quant_config, intermediate_size)
150 self.down_proj = RowParallelLinear(
151 intermediate_size,
vllm/model_executor/models/opt.py
-1
@@ -130,7 +130,6 @@ class OPTDecoderLayer(nn.Module):
130 bias=config.enable_bias,
131 quant_config=quant_config,
132 )
133 - quant_config = getattr(quant_config, "quant_config", None)
133 self.activation_fn = get_act_fn(config.activation_function,
134 quant_config, config.ffn_dim)
135 self.fc2 = RowParallelLinear(
vllm/model_executor/models/phi.py
-1
@@ -142,7 +142,6 @@ class PhiMLP(nn.Module):
142 config.hidden_size,
143 quant_config=quant_config,
144 )
145 - quant_config = getattr(quant_config, "quant_config", None)
145 self.act = get_act_fn(config.hidden_act, quant_config, n_inner)
146
147 def forward(self, hidden_states):
vllm/model_executor/models/starcoder2.py
-1
@@ -136,7 +136,6 @@ class Starcoder2MLP(nn.Module):
136 bias=config.use_bias,
137 quant_config=quant_config,
138 )
139 - quant_config = getattr(quant_config, "quant_config", None)
139 self.act = get_act_fn(config.hidden_act, quant_config,
140 config.intermediate_size)
141