Revert "add deepseek v2 support (casper-hansen#508)"

This reverts commit 6b45c95.
TechxGenus · Jun 27, 2024 · fe521ca · fe521ca
1 parent c53cc7e
commit fe521ca
Show file tree

Hide file tree

Showing 5 changed files with 1 addition and 140 deletions.
diff --git a/awq/models/__init__.py b/awq/models/__init__.py
@@ -19,5 +19,4 @@
 from .starcoder2 import Starcoder2AWQForCausalLM
 from .phi3 import Phi3AWQForCausalLM
 from .cohere import CohereAWQForCausalLM
-from .deepseek_v2 import DeepseekV2AWQForCausalLM
 from .minicpm import MiniCPMAWQForCausalLM
diff --git a/awq/models/auto.py b/awq/models/auto.py
@@ -28,7 +28,6 @@
  "starcoder2": Starcoder2AWQForCausalLM,
  "phi3": Phi3AWQForCausalLM,
  "cohere": CohereAWQForCausalLM,
- "deepseek_v2": DeepseekV2AWQForCausalLM,
  "minicpm": MiniCPMAWQForCausalLM,
 }
 

diff --git a/awq/models/base.py b/awq/models/base.py
@@ -80,8 +80,7 @@
  "starcoder2": "AutoModelForCausalLM",
  "phi3": "AutoModelForCausalLM",
  "cohere": "AutoModelForCausalLM",
- "deepseek_v2": "AutoModelForCausalLM",
- "minicpm":"AutoModelForCausalLM",
+ "minicpm":"AutoModelForCausalLM"
 }
 
 
@@ -507,8 +506,6 @@ def from_quantized(
  max_batch_size=int(os.getenv("AWQ_BATCH_SIZE", 1)),
  )
 
- model.eval()
-
  return self(
  model,
  model_type,

diff --git a/awq/models/deepseek_v2.py b/awq/models/deepseek_v2.py
diff --git a/awq/quantize/quantizer.py b/awq/quantize/quantizer.py
@@ -522,12 +522,6 @@ def cache_input_hook(m, x, y, name, feat_dict):
  "block_sparse_moe": layer.block_sparse_moe,
  }
 
- if self.awq_model.model_type == "deepseek_v2":
- named_linears = {
- **named_linears,
- "mlp": layer.mlp,
- }
-
  for name in named_linears:
  handles.append(
  named_linears[name].register_forward_hook(