Refactor llama family models (#2637)

This commit is contained in:
Roy
2024-02-13 00:09:23 -08:00
committed by GitHub
parent f964493274
commit 5c976a7e1a
17 changed files with 236 additions and 2720 deletions
+4 -5
View File
@@ -10,8 +10,8 @@ logger = init_logger(__name__)
# Architecture -> (module, class).
_MODELS = {
"AquilaModel": ("aquila", "AquilaForCausalLM"),
"AquilaForCausalLM": ("aquila", "AquilaForCausalLM"), # AquilaChat2
"AquilaModel": ("llama", "LlamaForCausalLM"),
"AquilaForCausalLM": ("llama", "LlamaForCausalLM"), # AquilaChat2
"BaiChuanForCausalLM": ("baichuan", "BaiChuanForCausalLM"), # baichuan-7b
"BaichuanForCausalLM": ("baichuan", "BaichuanForCausalLM"), # baichuan-13b
"BloomForCausalLM": ("bloom", "BloomForCausalLM"),
@@ -24,12 +24,12 @@ _MODELS = {
"GPTBigCodeForCausalLM": ("gpt_bigcode", "GPTBigCodeForCausalLM"),
"GPTJForCausalLM": ("gpt_j", "GPTJForCausalLM"),
"GPTNeoXForCausalLM": ("gpt_neox", "GPTNeoXForCausalLM"),
"InternLMForCausalLM": ("internlm", "InternLMForCausalLM"),
"InternLMForCausalLM": ("llama", "LlamaForCausalLM"),
"InternLM2ForCausalLM": ("internlm2", "InternLM2ForCausalLM"),
"LlamaForCausalLM": ("llama", "LlamaForCausalLM"),
# For decapoda-research/llama-*
"LLaMAForCausalLM": ("llama", "LlamaForCausalLM"),
"MistralForCausalLM": ("mistral", "MistralForCausalLM"),
"MistralForCausalLM": ("llama", "LlamaForCausalLM"),
"MixtralForCausalLM": ("mixtral", "MixtralForCausalLM"),
"QuantMixtralForCausalLM": ("mixtral_quant", "MixtralForCausalLM"),
# transformers's mpt class has lower case
@@ -41,7 +41,6 @@ _MODELS = {
"Qwen2ForCausalLM": ("qwen2", "Qwen2ForCausalLM"),
"RWForCausalLM": ("falcon", "FalconForCausalLM"),
"StableLMEpochForCausalLM": ("stablelm", "StablelmForCausalLM"),
"YiForCausalLM": ("yi", "YiForCausalLM")
}
# Models not supported by ROCm.