Files

T

ChayenneandChayenne 539df95d2c Imporve openai api documents (#1827 )

Co-authored-by: Chayenne <zhaochenyang@g.ucla.edu>

2024-10-30 00:39:41 -07:00

14 KiB

Raw Blame History

Embedding Model

SGLang supports embedding models in the same way as completion models. Here are some example models:

Launch A Server

The following code is equivalent to running this in the shell:

python -m sglang.launch_server --model-path Alibaba-NLP/gte-Qwen2-7B-instruct \
    --port 30010 --host 0.0.0.0 --is-embedding

Remember to add --is-embedding to the command.

In [1]:

from sglang.utils import (
    execute_shell_command,
    wait_for_server,
    terminate_process,
    print_highlight,
)

embedding_process = execute_shell_command(
    """
python -m sglang.launch_server --model-path Alibaba-NLP/gte-Qwen2-7B-instruct \
    --port 30010 --host 0.0.0.0 --is-embedding
"""
)

wait_for_server("http://localhost:30010")

/home/chenyang/miniconda3/envs/AlphaMeemory/lib/python3.11/site-packages/transformers/utils/hub.py:128: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
  warnings.warn(
[2024-10-29 21:07:15] server_args=ServerArgs(model_path='Alibaba-NLP/gte-Qwen2-7B-instruct', tokenizer_path='Alibaba-NLP/gte-Qwen2-7B-instruct', tokenizer_mode='auto', skip_tokenizer_init=False, load_format='auto', trust_remote_code=False, dtype='auto', kv_cache_dtype='auto', quantization=None, context_length=None, device='cuda', served_model_name='Alibaba-NLP/gte-Qwen2-7B-instruct', chat_template=None, is_embedding=True, host='0.0.0.0', port=30010, mem_fraction_static=0.88, max_running_requests=None, max_total_tokens=None, chunked_prefill_size=8192, max_prefill_tokens=16384, schedule_policy='lpm', schedule_conservativeness=1.0, tp_size=1, stream_interval=1, random_seed=568040040, constrained_json_whitespace_pattern=None, log_level='info', log_level_http=None, log_requests=False, show_time_cost=False, api_key=None, file_storage_pth='SGLang_storage', enable_cache_report=False, watchdog_timeout=600, dp_size=1, load_balance_method='round_robin', dist_init_addr=None, nnodes=1, node_rank=0, json_model_override_args='{}', enable_double_sparsity=False, ds_channel_config_path=None, ds_heavy_channel_num=32, ds_heavy_token_num=256, ds_heavy_channel_type='qk', ds_sparse_decode_threshold=4096, lora_paths=None, max_loras_per_batch=8, attention_backend='flashinfer', sampling_backend='flashinfer', grammar_backend='outlines', disable_flashinfer=False, disable_flashinfer_sampling=False, disable_radix_cache=False, disable_regex_jump_forward=False, disable_cuda_graph=False, disable_cuda_graph_padding=False, disable_disk_cache=False, disable_custom_all_reduce=False, disable_mla=False, disable_penalizer=False, disable_nan_detection=False, enable_overlap_schedule=False, enable_mixed_chunk=False, enable_torch_compile=False, torch_compile_max_bs=32, cuda_graph_max_bs=160, torchao_config='', enable_p2p_check=False, triton_attention_reduce_in_fp32=False, num_continuous_decode_steps=1)
/home/chenyang/miniconda3/envs/AlphaMeemory/lib/python3.11/site-packages/transformers/utils/hub.py:128: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
  warnings.warn(
/home/chenyang/miniconda3/envs/AlphaMeemory/lib/python3.11/site-packages/transformers/utils/hub.py:128: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
  warnings.warn(
[2024-10-29 21:07:20 TP0] Init torch distributed begin.
[2024-10-29 21:07:20 TP0] Load weight begin. avail mem=47.27 GB
[2024-10-29 21:07:21 TP0] lm_eval is not installed, GPTQ may not be usable
INFO 10-29 21:07:22 weight_utils.py:243] Using model weights format ['*.safetensors']
Loading safetensors checkpoint shards:   0% Completed | 0/7 [00:00<?, ?it/s]
Loading safetensors checkpoint shards:  14% Completed | 1/7 [00:00<00:03,  1.65it/s]
Loading safetensors checkpoint shards:  29% Completed | 2/7 [00:01<00:04,  1.02it/s]
Loading safetensors checkpoint shards:  43% Completed | 3/7 [00:03<00:04,  1.24s/it]
Loading safetensors checkpoint shards:  57% Completed | 4/7 [00:05<00:04,  1.47s/it]
Loading safetensors checkpoint shards:  71% Completed | 5/7 [00:07<00:03,  1.62s/it]
Loading safetensors checkpoint shards:  86% Completed | 6/7 [00:08<00:01,  1.64s/it]
Loading safetensors checkpoint shards: 100% Completed | 7/7 [00:10<00:00,  1.63s/it]
Loading safetensors checkpoint shards: 100% Completed | 7/7 [00:10<00:00,  1.49s/it]

[2024-10-29 21:07:32 TP0] Load weight end. type=Qwen2ForCausalLM, dtype=torch.float16, avail mem=32.91 GB
[2024-10-29 21:07:33 TP0] Memory pool end. avail mem=4.56 GB
[2024-10-29 21:07:33 TP0] max_total_num_tokens=509971, max_prefill_tokens=16384, max_running_requests=2049, context_len=131072
[2024-10-29 21:07:33] INFO:     Started server process [2650986]
[2024-10-29 21:07:33] INFO:     Waiting for application startup.
[2024-10-29 21:07:33] INFO:     Application startup complete.
[2024-10-29 21:07:33] INFO:     Uvicorn running on http://0.0.0.0:30010 (Press CTRL+C to quit)
[2024-10-29 21:07:34] INFO:     127.0.0.1:47812 - "GET /v1/models HTTP/1.1" 200 OK

This cell combines server and notebook output.

Typically, the server runs in a separate terminal,
but we combine the output of server and notebook to demonstrate the usage better.

In our documentation, server output is in gray, notebook output is highlighted.

[2024-10-29 21:07:34] INFO:     127.0.0.1:41780 - "GET /get_model_info HTTP/1.1" 200 OK
[2024-10-29 21:07:34 TP0] Prefill batch. #new-seq: 1, #new-token: 6, #cached-token: 0, cache hit rate: 0.00%, token usage: 0.00, #running-req: 0, #queue-req: 0
[2024-10-29 21:07:35] INFO:     127.0.0.1:41792 - "POST /encode HTTP/1.1" 200 OK
[2024-10-29 21:07:35] The server is fired up and ready to roll!

Use Curl

In [2]:

import subprocess, json

text = "Once upon a time"

curl_text = f"""curl -s http://localhost:30010/v1/embeddings \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer None" \
  -d '{{"model": "Alibaba-NLP/gte-Qwen2-7B-instruct", "input": "{text}"}}'"""

text_embedding = json.loads(subprocess.check_output(curl_text, shell=True))["data"][0][
    "embedding"
]

print_highlight(f"Text embedding (first 10): {text_embedding[:10]}")

[2024-10-28 02:10:30 TP0] Prefill batch. #new-seq: 1, #new-token: 4, #cached-token: 0, cache hit rate: 0.00%, token usage: 0.00, #running-req: 0, #queue-req: 0

[2024-10-28 02:10:31] INFO:     127.0.0.1:48094 - "POST /v1/embeddings HTTP/1.1" 200 OK

Text embedding (first 10): [0.0083160400390625, 0.0006804466247558594, -0.00809478759765625, -0.0006995201110839844, 0.0143890380859375, -0.0090179443359375, 0.01238250732421875, 0.00209808349609375, 0.0062103271484375, -0.003047943115234375]

Using OpenAI Compatible API

In [3]:

import openai

client = openai.Client(base_url="http://127.0.0.1:30010/v1", api_key="None")

# Text embedding example
response = client.embeddings.create(
    model="Alibaba-NLP/gte-Qwen2-7B-instruct",
    input=text,
)

embedding = response.data[0].embedding[:10]
print_highlight(f"Text embedding (first 10): {embedding}")

[2024-10-28 02:10:31] INFO:     127.0.0.1:48110 - "GET /get_model_info HTTP/1.1" 200 OK
[2024-10-28 02:10:31 TP0] Prefill batch. #new-seq: 1, #new-token: 6, #cached-token: 0, cache hit rate: 0.00%, token usage: 0.00, #running-req: 0, #queue-req: 0
[2024-10-28 02:10:31] INFO:     127.0.0.1:48114 - "POST /encode HTTP/1.1" 200 OK
[2024-10-28 02:10:31] The server is fired up and ready to roll!
[2024-10-28 02:10:31 TP0] Prefill batch. #new-seq: 1, #new-token: 1, #cached-token: 3, cache hit rate: 21.43%, token usage: 0.00, #running-req: 0, #queue-req: 0
[2024-10-28 02:10:31] INFO:     127.0.0.1:48118 - "POST /v1/embeddings HTTP/1.1" 200 OK

Text embedding (first 10): [0.00829315185546875, 0.0007004737854003906, -0.00809478759765625, -0.0006799697875976562, 0.01438140869140625, -0.00897979736328125, 0.0123748779296875, 0.0020923614501953125, 0.006195068359375, -0.0030498504638671875]

Using Input IDs

SGLang also supports input_ids as input to get the embedding.

In [4]:

import json
import os
from transformers import AutoTokenizer

os.environ["TOKENIZERS_PARALLELISM"] = "false"

tokenizer = AutoTokenizer.from_pretrained("Alibaba-NLP/gte-Qwen2-7B-instruct")
input_ids = tokenizer.encode(text)

curl_ids = f"""curl -s http://localhost:30010/v1/embeddings \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer None" \
  -d '{{"model": "Alibaba-NLP/gte-Qwen2-7B-instruct", "input": {json.dumps(input_ids)}}}'"""

input_ids_embedding = json.loads(subprocess.check_output(curl_ids, shell=True))["data"][
    0
]["embedding"]

print_highlight(f"Input IDs embedding (first 10): {input_ids_embedding[:10]}")

/home/chenyang/miniconda3/envs/AlphaMeemory/lib/python3.11/site-packages/transformers/utils/hub.py:127: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
  warnings.warn(

[2024-10-28 02:10:32 TP0] Prefill batch. #new-seq: 1, #new-token: 1, #cached-token: 3, cache hit rate: 33.33%, token usage: 0.00, #running-req: 0, #queue-req: 0
[2024-10-28 02:10:32] INFO:     127.0.0.1:48124 - "POST /v1/embeddings HTTP/1.1" 200 OK

Input IDs embedding (first 10): [0.00829315185546875, 0.0007004737854003906, -0.00809478759765625, -0.0006799697875976562, 0.01438140869140625, -0.00897979736328125, 0.0123748779296875, 0.0020923614501953125, 0.006195068359375, -0.0030498504638671875]

In [5]:

terminate_process(embedding_process)

[2024-10-28 02:10:32] INFO:     Shutting down
[2024-10-28 02:10:32] INFO:     Waiting for application shutdown.
[2024-10-28 02:10:32] INFO:     Application shutdown complete.
[2024-10-28 02:10:32] INFO:     Finished server process [1188896]
W1028 02:10:32.490000 140389363193408 torch/_inductor/compile_worker/subproc_pool.py:126] SubprocPool unclean exit

14 KiB Raw Blame History