tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

support to disable exllama for gptq (#604)

Browse Source 
* support to disable exllama for gptq

* update property instead of item

* fix config key
This commit is contained in:
Wing Lian
2023-09-19 17:51:08 -04:00
committed by GitHub
parent aa656e04bd
commit faecff9798
2 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
examples/llama-2/gptq-lora.yml
View File

@@ -2,7 +2,7 @@ base_model: TheBloke/Llama-2-7B-GPTQ
base_model_config: TheBloke/Llama-2-7B-GPTQ base_model_config: TheBloke/Llama-2-7B-GPTQ
is_llama_derived_model: false is_llama_derived_model: false
gptq: true gptq: true
gptq_bits: 4 gptq_disable_exllama: true
model_type: AutoModelForCausalLM model_type: AutoModelForCausalLM
tokenizer_type: LlamaTokenizer tokenizer_type: LlamaTokenizer
tokenizer_use_fast: true tokenizer_use_fast: true
@@ -62,8 +62,6 @@ xformers_attention:
flash_attention: flash_attention:
sdp_attention: sdp_attention:
flash_optimum: flash_optimum:
gptq_groupsize:
gptq_model_v1:
warmup_steps: 100 warmup_steps: 100
eval_steps: eval_steps:
save_steps: save_steps:

4
src/axolotl/utils/models.py
View File

@@ -196,6 +196,10 @@ def load_model(
if not hasattr(model_config, "quantization_config"): if not hasattr(model_config, "quantization_config"):
LOG.warning("model config does not contain quantization_config information") LOG.warning("model config does not contain quantization_config information")
else: else:
if cfg.gptq_disable_exllama is not None:
model_config.quantization_config[
"disable_exllama"
] = cfg.gptq_disable_exllama
model_kwargs["quantization_config"] = GPTQConfig( model_kwargs["quantization_config"] = GPTQConfig(
**model_config.quantization_config **model_config.quantization_config
) )