Skip to content

Support mixtral kvcache reuse and remove kv_cache_fp8 #898

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
regisss merged 8 commits into from
May 4, 2024
Merged

Support mixtral kvcache reuse and remove kv_cache_fp8 #898

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
2394afd
Add linear and dynamic RoPE to Mistral and Mixtral
regisss Apr 16, 2024
b004ff3
Add references
regisss Apr 16, 2024
43d194d
Support mixtral kv-cache reuse and remove kv_cache_fp8
Apr 17, 2024
a05b986
Modify white/black list to allow/block list
Apr 17, 2024
af8e9cd
add mixtral fp8 test case
Apr 23, 2024
07aba12
merge main branch
Apr 28, 2024
7ca1876
fix merge conflict
Apr 28, 2024
07572a7
Merge branch 'main' into support-mixtral-kvcache-reuse
regisss May 4, 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
4 changes: 2 additions & 2 deletions examples/text-generation/quantization_config/maxabs_quant_mixtral.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,8 +3,8 @@
"mode": "QUANTIZE",
"observer": "maxabs",
"scale_method": "maxabs_hw",
"whitelist": {"types": [], "names": ["gate","w1","w3","w2"]},
"blacklist": {"types": [], "names": [
"allowlist": {"types": [], "names": ["gate","w1","w3","w2"]},
"blocklist": {"types": [], "names": [
"model.layers.1.block_sparse_moe.experts.(3|4).w2",
"model.layers.[29-31].block_sparse_moe.experts.[0-7].w2"
]},
Expand Down
3 changes: 2 additions & 1 deletion optimum/habana/transformers/generation/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -604,7 +604,8 @@ def generate(
"llama",
"mistral",
"falcon",
], "reuse_cache only supported by llama, mistral and falcon at the moment"
"mixtral",
], "reuse_cache only supported by llama, mistral, falcon and mixtral at the moment"
if not generation_config.bucket_internal:
assert (
generation_config.bucket_size <= 0
Expand Down
4 changes: 2 additions & 2 deletions optimum/habana/transformers/modeling_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,7 @@
GaudiMixtralAttention,
GaudiMixtralDecoderLayer,
GaudiMixtralForCausalLM,
GaudiMixtralModel,
GaudiMptForCausalLM,
GaudiMptModel,
GaudiOPTForCausalLM,
Expand Down Expand Up @@ -120,7 +121,6 @@
gaudi_llama_rmsnorm_forward,
gaudi_mistral_rmsnorm_forward,
gaudi_mixtral_block_sparse_moe_forward,
gaudi_mixtral_model_forward,
gaudi_mixtral_rmsnorm_forward,
gaudi_mpt_attention_forward,
gaudi_mpt_block_forward,
Expand Down Expand Up @@ -392,7 +392,7 @@ def adapt_transformers_to_gaudi():
# Optimization for mixtral on Gaudi
transformers.models.mixtral.modeling_mixtral.MixtralAttention = GaudiMixtralAttention
transformers.models.mixtral.modeling_mixtral.MixtralForCausalLM = GaudiMixtralForCausalLM
transformers.models.mixtral.modeling_mixtral.MixtralModel.forward = gaudi_mixtral_model_forward
transformers.models.mixtral.modeling_mixtral.MixtralModel = GaudiMixtralModel
transformers.models.mixtral.modeling_mixtral.MixtralSparseMoeBlock.forward = gaudi_mixtral_block_sparse_moe_forward
transformers.models.mixtral.modeling_mixtral.MixtralDecoderLayer = GaudiMixtralDecoderLayer
transformers.models.mixtral.modeling_mixtral.MixtralRMSNorm.forward = gaudi_mixtral_rmsnorm_forward
Expand Down
2 changes: 1 addition & 1 deletion optimum/habana/transformers/models/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -101,9 +101,9 @@
GaudiMixtralAttention,
GaudiMixtralDecoderLayer,
GaudiMixtralForCausalLM,
GaudiMixtralModel,
MixtralConfig,
gaudi_mixtral_block_sparse_moe_forward,
gaudi_mixtral_model_forward,
gaudi_mixtral_rmsnorm_forward,
)
from .modeling_all_models import (
Expand Down
2 changes: 1 addition & 1 deletion optimum/habana/transformers/models/mixtral/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
GaudiMixtralAttention,
GaudiMixtralDecoderLayer,
GaudiMixtralForCausalLM,
GaudiMixtralModel,
gaudi_mixtral_block_sparse_moe_forward,
gaudi_mixtral_model_forward,
gaudi_mixtral_rmsnorm_forward,
)
Loading