Add LoRA support for Mixtral (#2831)

* add mixtral lora support

* formatting

* fix incorrectly ported logic

* polish tests

* minor fixes and refactoring

* minor fixes

* formatting

* rename and remove redundant logic

* refactoring

* refactoring

* minor fix

* minor refactoring

* fix code smell
This commit is contained in:
Terry
2024-02-14 00:55:45 +01:00
committed by GitHub
parent 317b29de0f
commit 2a543d6efe
10 changed files with 251 additions and 121 deletions
+10 -1
View File
@@ -86,11 +86,20 @@ class ModelRunner:
vocab_size = self.model.config.vocab_size
if self.lora_config:
assert hasattr(
self.model, "supported_lora_modules"
) and self.model.supported_lora_modules, "Model does not support LoRA"
assert hasattr(
self.model,
"embedding_modules"), "Model does not have embedding_modules"
assert hasattr(self.model, "embedding_padding_modules"
), "Model does not have embedding_padding_modules"
self.lora_manager = LRUCacheWorkerLoRAManager(
self.scheduler_config.max_num_seqs,
self.scheduler_config.max_num_batched_tokens +
self.scheduler_config.max_paddings, vocab_size,
self.lora_config, self.device)
self.lora_config, self.device, self.model.embedding_modules,
self.model.embedding_padding_modules)
self.model = self.lora_manager.create_lora_manager(self.model)
def set_block_size(self, block_size: int) -> None: