mirror of
https://github.com/huggingface/text-generation-inference.git
synced 2025-04-22 15:32:08 +00:00
7e2a7433d3
This PR adds basic modeling for phi-2
run
```bash
text-generation-server \
serve \
microsoft/phi-2 \
--revision 834565c23f9b28b96ccbeabe614dd906b6db551a
```
test
```bash
curl -s localhost:3000/generate \
-X POST \
-d '{"inputs":"What is Deep Learning?","parameters":{"max_new_tokens":20}}' \
-H 'Content-Type: application/json' | jq .
# {
# "generated_text": "\nDeep learning is a subset of machine learning that uses artificial neural networks to learn from data. These"
# }
```
notes
- recently (~1 day ago) the Phi weights and model were updated to
accommodate adding [GQA/MQA attention to the
model.](https://github.com/huggingface/transformers/pull/28163) This
impl expects the original model format so a fixed revision is required
at the moment.
- this PR only includes a basic implementation of the model and can
later be extended for support Flash and Sharded versions as well as make
use of better optimization
64 lines
2.1 KiB
Python
64 lines
2.1 KiB
Python
import torch
|
|
import torch.distributed
|
|
|
|
from transformers import AutoConfig, AutoTokenizer
|
|
from typing import Optional, List, Tuple
|
|
|
|
from text_generation_server.models import CausalLM
|
|
from text_generation_server.models.custom_modeling.phi_modeling import PhiConfig, PhiForCausalLM
|
|
from text_generation_server.utils import (
|
|
initialize_torch_distributed,
|
|
weight_files,
|
|
Weights,
|
|
)
|
|
|
|
class Phi(CausalLM):
|
|
def __init__(
|
|
self,
|
|
model_id: str,
|
|
revision: Optional[str] = None,
|
|
quantize: Optional[str] = None,
|
|
dtype: Optional[torch.dtype] = None,
|
|
trust_remote_code: bool = False,
|
|
):
|
|
self.process_group, _rank, _world_size = initialize_torch_distributed()
|
|
if torch.cuda.is_available():
|
|
device = torch.device("cuda")
|
|
dtype = torch.float16 if dtype is None else dtype
|
|
else:
|
|
if quantize:
|
|
raise ValueError("quantization is not available on CPU")
|
|
|
|
device = torch.device("cpu")
|
|
dtype = torch.float32 if dtype is None else dtype
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(
|
|
model_id,
|
|
revision=revision,
|
|
padding_side="left",
|
|
truncation_side="left",
|
|
trust_remote_code=trust_remote_code,
|
|
)
|
|
config = PhiConfig.from_pretrained(
|
|
model_id, revision=revision, trust_remote_code=trust_remote_code
|
|
)
|
|
|
|
tokenizer.bos_token_id = config.bos_token_id
|
|
tokenizer.eos_token_id = config.eos_token_id
|
|
tokenizer.pad_token = tokenizer.eos_token
|
|
|
|
config.quantize = quantize
|
|
torch.distributed.barrier(group=self.process_group)
|
|
filenames = weight_files(model_id, revision=revision, extension=".safetensors")
|
|
weights = Weights(filenames, device, dtype, process_group=self.process_group)
|
|
model = PhiForCausalLM(config, weights)
|
|
torch.distributed.barrier(group=self.process_group)
|
|
super(CausalLM, self).__init__(
|
|
model=model,
|
|
tokenizer=tokenizer,
|
|
requires_padding=True,
|
|
dtype=dtype,
|
|
device=device,
|
|
)
|
|
|