ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

revert

Browse source
This commit is contained in:
Daniel Han-Chen 2024-02-24 18:25:25 +11:00
commit 82e45c02a5
2 changed files with 5 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

8
unsloth/models/gemma.py
View file

@ -582,10 +582,10 @@ class FastGemmaModel(FastLlamaModel):
# GemmaAttention .forward = GemmaAttention_fast_forward
# GemmaSdpaAttention .forward = GemmaAttention_fast_forward
# GemmaFlashAttention2.forward = GemmaAttention_fast_forward
GemmaDecoderLayer .forward = GemmaDecoderLayer_fast_forward
GemmaModel .forward = GemmaModel_fast_forward
GemmaForCausalLM .forward = GemmaForCausalLM_fast_forward
PeftModelForCausalLM.forward = PeftModelForCausalLM_fast_forward
# GemmaDecoderLayer .forward = GemmaDecoderLayer_fast_forward
# GemmaModel .forward = GemmaModel_fast_forward
# GemmaForCausalLM .forward = GemmaForCausalLM_fast_forward
# PeftModelForCausalLM.forward = PeftModelForCausalLM_fast_forward
# Solves https://github.com/unslothai/unsloth/issues/168
# Static KV Cache was introduced in 4.38.0, causing training to be much slower.

2
unsloth/models/llama.py
View file

@ -1400,7 +1400,7 @@ class FastLlamaModel:
(down_proj.base_layer if hasattr(down_proj, "base_layer") else down_proj).bias is None:
# https://stackoverflow.com/questions/50599045/python-replacing-a-function-within-a-class-of-a-module
layer.mlp.forward = types.MethodType(apply_lora_mlp, layer.mlp)
# layer.mlp.forward = types.MethodType(apply_lora_mlp, layer.mlp)
n_mlp += 1
else:
logger.warning_once(