Skip to content
This repository has been archived by the owner on Oct 25, 2024. It is now read-only.

Support lmhead int4 #1670

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Support lmhead int4 #1670

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 6 additions & 5 deletions intel_extension_for_transformers/transformers/llm/quantization/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -136,7 +136,8 @@ def replace_linear(
if modules_to_not_convert is None:
# output_layer is chatglm last layer name
# embed_out is dolly_v2 last layer name
modules_to_not_convert = ["lm_head", "output_layer", "embed_out"]
#modules_to_not_convert = ["lm_head", "output_layer", "embed_out"]
modules_to_not_convert = []
if quantization_config.llm_int8_skip_modules:
modules_to_not_convert = modules_to_not_convert.extend(
quantization_config.llm_int8_skip_modules
Expand Down Expand Up @@ -662,10 +663,10 @@ def convert_to_quantized_model(model, config, device="cpu"):
iters=config.iters,
scale_dtype=config.scale_dtype,
)
if config.quant_lm_head is False:
quant_config.set_local(".*lm_head", AutoRoundConfig(dtype="fp32"))
quant_config.set_local(".*output_layer", AutoRoundConfig(dtype="fp32"))
quant_config.set_local(".*embed_out", AutoRoundConfig(dtype="fp32"))
#if config.quant_lm_head is False:
# quant_config.set_local(".*lm_head", AutoRoundConfig(dtype="fp32"))
# quant_config.set_local(".*output_layer", AutoRoundConfig(dtype="fp32"))
# quant_config.set_local(".*embed_out", AutoRoundConfig(dtype="fp32"))
logger.info(f"Do AutoRound algorithm with config {quant_config}")
dataloader = get_autoround_dataloader(tokenizer=config.tokenizer,
seqlen=config.seq_len,
Expand Down
Loading