ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

add compresskv back for mistral (#12607)

Browse source 
* add compresskv back for mistral

* fix

* fix
This commit is contained in:
Yishuo Wang 2024-12-25 11:06:08 +08:00 committed by GitHub
parent 9c9800be31
commit 4e6b9d804f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 29 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
python/llm/src/ipex_llm/transformers/models/mistral.py
View file

@ -37,6 +37,7 @@
from typing import Optional, Tuple, Union, List from typing import Optional, Tuple, Union, List
import os
import torch import torch
from transformers.cache_utils import Cache from transformers.cache_utils import Cache
from transformers.modeling_outputs import BaseModelOutputWithPast from transformers.modeling_outputs import BaseModelOutputWithPast
@ -45,8 +46,11 @@ from transformers.models.mistral.modeling_mistral import MistralModel, MistralAt
from ipex_llm.transformers.models.common import merge_qkv_base from ipex_llm.transformers.models.common import merge_qkv_base
from ipex_llm.transformers.models.common import scaled_dot_product_attention from ipex_llm.transformers.models.common import scaled_dot_product_attention
from ipex_llm.transformers.models.utils import should_use_fuse_rope, apply_rotary_pos_emb from ipex_llm.transformers.models.utils import should_use_fuse_rope, apply_rotary_pos_emb
from ipex_llm.transformers.models.utils import should_use_compresskv, is_enough_kv_cache_room_4_36
from ipex_llm.transformers.models.utils import use_quantize_kv_cache from ipex_llm.transformers.models.utils import use_quantize_kv_cache
from ipex_llm.transformers.kv import DynamicFp8Cache, DynamicNormalCache from ipex_llm.transformers.kv import DynamicFp8Cache, DynamicNormalCache
from ipex_llm.transformers.kv import DynamicCompressCache, DynamicCompressFp8Cache
KV_CACHE_ALLOC_BLOCK_LENGTH = int(os.environ.get("KV_CACHE_ALLOC_BLOCK_LENGTH", 256))
def mistral_model_forward( def mistral_model_forward(
@ -69,11 +73,22 @@ def mistral_model_forward(
use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.down_proj, inputs, use_quantize_kv = use_quantize_kv_cache(self.layers[0].mlp.down_proj, inputs,
self.config.num_attention_heads // self.config.num_attention_heads //
self.config.num_key_value_heads) self.config.num_key_value_heads)
use_compress_kv = should_use_compresskv(inputs, inputs.size(1)) or \
isinstance(past_key_values, DynamicCompressCache)
if use_cache: if use_cache:
if use_quantize_kv and not isinstance(past_key_values, DynamicFp8Cache): if use_compress_kv and not isinstance(past_key_values, DynamicCompressCache):
if use_quantize_kv:
past_key_values = DynamicCompressFp8Cache.from_legacy_cache(past_key_values)
else:
past_key_values = DynamicCompressCache.from_legacy_cache(past_key_values)
elif use_quantize_kv and not isinstance(past_key_values, DynamicFp8Cache):
past_key_values = DynamicFp8Cache.from_legacy_cache(past_key_values) past_key_values = DynamicFp8Cache.from_legacy_cache(past_key_values)
elif not use_quantize_kv and not isinstance(past_key_values, DynamicNormalCache): elif (
not use_quantize_kv
and not use_compress_kv
and not isinstance(past_key_values, DynamicNormalCache)
):
past_key_values = DynamicNormalCache.from_legacy_cache(past_key_values) past_key_values = DynamicNormalCache.from_legacy_cache(past_key_values)
# ipex-llm changes end # ipex-llm changes end
@ -127,8 +142,16 @@ def mistral_attention_forward(
query_states, key_states = apply_rotary_pos_emb(query_states, key_states, query_states, key_states = apply_rotary_pos_emb(query_states, key_states,
cos, sin, position_ids, "mistral") cos, sin, position_ids, "mistral")
key_states, value_states = past_key_value.update(key_states, value_states, if isinstance(past_key_value, DynamicCompressCache):
self.layer_idx, None) enough_kv_room = is_enough_kv_cache_room_4_36(past_key_value, self.layer_idx, q_len)
key_states, value_states = past_key_value.update(
key_states, value_states, self.layer_idx,
query_states, attention_mask, self.num_key_value_groups,
self.config, enough_kv_room, KV_CACHE_ALLOC_BLOCK_LENGTH
)
else:
key_states, value_states = past_key_value.update(key_states, value_states,
self.layer_idx, None)
# IPEX-LLM OPT: sdpa # IPEX-LLM OPT: sdpa
attn_weights = None attn_weights = None

4
python/llm/src/ipex_llm/transformers/models/mixtral.py
View file

@ -52,7 +52,7 @@ from ipex_llm.ggml.quantize import ggml_tensor_qtype
from ipex_llm.utils.common import invalidInputError from ipex_llm.utils.common import invalidInputError
from ipex_llm.transformers.models.utils import init_kv_cache, extend_kv_cache, append_kv_cache from ipex_llm.transformers.models.utils import init_kv_cache, extend_kv_cache, append_kv_cache
from ipex_llm.transformers.models.utils import apply_rotary_pos_emb, is_enough_kv_cache_room_4_36 from ipex_llm.transformers.models.utils import apply_rotary_pos_emb, is_enough_kv_cache_room_4_36
from ipex_llm.transformers.models.mistral import should_use_fuse_rope from ipex_llm.transformers.models.utils import should_use_fuse_rope
from ipex_llm.transformers.models.utils import use_decoding_fast_path from ipex_llm.transformers.models.utils import use_decoding_fast_path
from ipex_llm.transformers.models.utils import use_flash_attention, use_sdp from ipex_llm.transformers.models.utils import use_flash_attention, use_sdp
from ipex_llm.transformers.models.utils import mlp_fusion_check, SILU from ipex_llm.transformers.models.utils import mlp_fusion_check, SILU
@ -171,7 +171,7 @@ def mixtral_attention_forward(
# for flash attention # for flash attention
original_dtype = hidden_states.dtype original_dtype = hidden_states.dtype
use_fuse_rope = should_use_fuse_rope(self, hidden_states, position_ids) use_fuse_rope = should_use_fuse_rope(hidden_states, position_ids, self.training)
enough_kv_room = is_enough_kv_cache_room_4_36(past_key_value, self.layer_idx) enough_kv_room = is_enough_kv_cache_room_4_36(past_key_value, self.layer_idx)
decoding_fast_path = use_decoding_fast_path(self.q_proj, decoding_fast_path = use_decoding_fast_path(self.q_proj,
use_fuse_rope, use_fuse_rope,