[Cleanup] Remove online quantization support in fp8.py in favor of online shorthands (#53585)
Signed-off-by: Felix Marty <Felix.Marty@amd.com> Signed-off-by: mgoin <mgoin64@gmail.com> Co-authored-by: mgoin <mgoin64@gmail.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
mgoin
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
58aa2f1ef1
commit
90a9515006
@@ -796,11 +796,11 @@ def make_quant_config(
|
||||
return None, QuantizedWeights(w13_weight=w1, w2_weight=w2)
|
||||
|
||||
if quantization == "fp8":
|
||||
return Fp8Config(True), _quantize_fp8_halves(w1, w2)
|
||||
return Fp8Config(), _quantize_fp8_halves(w1, w2)
|
||||
|
||||
if quantization == "fp8_blocked":
|
||||
block_shape = [128, 128]
|
||||
return Fp8Config(True, weight_block_size=block_shape), _quantize_fp8_halves(
|
||||
return Fp8Config(weight_block_size=block_shape), _quantize_fp8_halves(
|
||||
w1, w2, block_shape
|
||||
)
|
||||
|
||||
|
||||
@@ -641,7 +641,7 @@ def make_shared_experts_with_weights(
|
||||
if quant_dtype == torch.float8_e4m3fn:
|
||||
from vllm.model_executor.layers.quantization.fp8 import Fp8Config
|
||||
|
||||
quant_config = Fp8Config(True)
|
||||
quant_config = Fp8Config()
|
||||
else:
|
||||
quant_config = None
|
||||
|
||||
|
||||
Reference in New Issue
Block a user