Mirrors
/
hf_text-generation-inference
mirror of https://github.com/huggingface/text-generation-inference.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix(server): Avoid using try/except to determine kind of AutoModel (#142)

This commit is contained in:
Nick Hill 2023-03-27 00:23:22 -07:00 committed by GitHub
parent ab5fd8cf93
commit 462530c2b0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 10 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
server/text_generation_server/models/__init__.py
View File

@ -3,6 +3,7 @@ import torch
from loguru import logger from loguru import logger
from transformers import AutoConfig from transformers import AutoConfig
from transformers.models.auto import modeling_auto
from typing import Optional from typing import Optional
from text_generation_server.models.model import Model from text_generation_server.models.model import Model
@ -65,14 +66,15 @@ def get_model(
return SantaCoder(model_id, revision, quantize) return SantaCoder(model_id, revision, quantize)
config = AutoConfig.from_pretrained(model_id, revision=revision) config = AutoConfig.from_pretrained(model_id, revision=revision)
model_type = config.model_type
if config.model_type == "bloom": if model_type == "bloom":
if sharded: if sharded:
return BLOOMSharded(model_id, revision, quantize=quantize) return BLOOMSharded(model_id, revision, quantize=quantize)
else: else:
return BLOOM(model_id, revision, quantize=quantize) return BLOOM(model_id, revision, quantize=quantize)
if config.model_type == "gpt_neox": if model_type == "gpt_neox":
if sharded: if sharded:
neox_cls = FlashNeoXSharded if FLASH_NEOX else GPTNeoxSharded neox_cls = FlashNeoXSharded if FLASH_NEOX else GPTNeoxSharded
return neox_cls(model_id, revision, quantize=quantize) return neox_cls(model_id, revision, quantize=quantize)
@ -80,7 +82,7 @@ def get_model(
neox_cls = FlashNeoX if FLASH_NEOX else CausalLM neox_cls = FlashNeoX if FLASH_NEOX else CausalLM
return neox_cls(model_id, revision, quantize=quantize) return neox_cls(model_id, revision, quantize=quantize)
if config.model_type == "t5": if model_type == "t5":
if sharded: if sharded:
return T5Sharded(model_id, revision, quantize=quantize) return T5Sharded(model_id, revision, quantize=quantize)
else: else:
@ -88,7 +90,10 @@ def get_model(
if sharded: if sharded:
raise ValueError("sharded is not supported for AutoModel") raise ValueError("sharded is not supported for AutoModel")
try:
if model_type in modeling_auto.MODEL_FOR_CAUSAL_LM_MAPPING_NAMES:
return CausalLM(model_id, revision, quantize=quantize) return CausalLM(model_id, revision, quantize=quantize)
except Exception: if model_type in modeling_auto.MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES:
return Seq2SeqLM(model_id, revision, quantize=quantize) return Seq2SeqLM(model_id, revision, quantize=quantize)
raise ValueError(f"Unsupported model type {model_type}")