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This repository has been archived by the owner on Oct 25, 2024.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

fix int8 skip module config #1682

Merged
merged 7 commits into from
Aug 9, 2024
Merged

fix int8 skip module config #1682

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
ef2a1d4
fix int8 skip module config
changwangss Aug 5, 2024
0bc8428
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Aug 5, 2024
a08344a
fix embed_out
changwangss Aug 5, 2024
7914ccc
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Aug 5, 2024
1de8b17
Update modeling_auto.py
changwangss Aug 5, 2024
827c95d
Update config.py
changwangss Aug 6, 2024
439511c
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot] Aug 6, 2024
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
17 changes: 12 additions & 5 deletions intel_extension_for_transformers/transformers/utils/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -831,7 +831,10 @@ def __init__(
self.double_quant_bits = double_quant_bits
self.double_quant_use_sym = double_quant_use_sym
self.double_quant_group_size = double_quant_group_size
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules", ["lm_head", "output_layer", "embed_out"])
# "transformer.output_layer" for chatglm series model.
# "embed_out" for dolly v2 series model.
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules",
["lm_head", "transformer.output_layer", "embed_out"])
self.use_ggml = use_ggml
self.use_quant = use_quant
self.use_neural_speed = use_neural_speed
Expand Down Expand Up @@ -911,7 +914,8 @@ def __init__(
self.true_sequential = true_sequential
self.layer_wise = layer_wise
self.seq_len = seq_len
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules", ["lm_head", "output_layer", "embed_out"])
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules",
["lm_head", "transformer.output_layer", "embed_out"])
self.use_ggml = use_ggml
self.use_quant = use_quant
self.use_neural_speed = use_neural_speed
Expand Down Expand Up @@ -1009,7 +1013,8 @@ def __init__(
self.seq_len = seq_len
self.use_double_quant = use_double_quant
self.double_quant_scale_dtype = double_quant_scale_dtype
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules", ["lm_head", "output_layer", "embed_out"])
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules",
["lm_head", "transformer.output_layer", "embed_out"])
self.use_ggml = use_ggml
self.use_quant = use_quant
self.use_neural_speed = use_neural_speed
Expand Down Expand Up @@ -1078,7 +1083,8 @@ def __init__(
self.seq_len = seq_len
self.use_double_quant = use_double_quant
self.double_quant_scale_dtype = double_quant_scale_dtype
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules", ["lm_head", "output_layer", "embed_out"])
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules",
["lm_head", "transformer.output_layer", "embed_out"])
self.use_ggml = use_ggml
self.use_neural_speed = use_neural_speed
self.device = kwargs.get("device", "auto")
Expand Down Expand Up @@ -1154,7 +1160,8 @@ def __init__(
self.iters = iters
self.seq_len = seq_len
self.quant_lm_head = quant_lm_head
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules", ["lm_head", "output_layer", "embed_out"])
self.llm_int8_skip_modules = kwargs.get("llm_int8_skip_modules",
["lm_head", "transformer.output_layer", "embed_out"])
if self.quant_lm_head:
self.llm_int8_skip_modules = []
self.use_ggml = use_ggml
Expand Down
Loading