-
Notifications
You must be signed in to change notification settings - Fork 1.2k
/
hf_client.py
494 lines (413 loc) · 17.3 KB
/
hf_client.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
import os
import random
import re
import shutil
import subprocess
from typing import Literal
# from dsp.modules.adapter import TurboAdapter, DavinciAdapter, LlamaAdapter
import backoff
import requests
from dsp.modules.cache_utils import CacheMemory, NotebookCacheMemory
from dsp.modules.hf import HFModel, openai_to_hf
ERRORS = (Exception)
def backoff_hdlr(details):
"""Handler from https://pypi.org/project/backoff/"""
print(
"Backing off {wait:0.1f} seconds after {tries} tries "
"calling function {target} with kwargs "
"{kwargs}".format(**details),
)
class HFClientTGI(HFModel):
def __init__(self, model, port, url="http:https://future-hgx-1", http_request_kwargs=None, **kwargs):
super().__init__(model=model, is_client=True)
self.url = url
self.ports = port if isinstance(port, list) else [port]
self.http_request_kwargs = http_request_kwargs or {}
self.headers = {"Content-Type": "application/json"}
self.kwargs = {
"model": model,
"port": port,
"url": url,
"temperature": 0.01,
"max_tokens": 75,
"top_p": 0.97,
"n": 1,
"stop": ["\n", "\n\n"],
**kwargs,
}
# print(self.kwargs)
def _generate(self, prompt, **kwargs):
kwargs = {**self.kwargs, **kwargs}
payload = {
"inputs": prompt,
"parameters": {
"do_sample": kwargs["n"] > 1,
"best_of": kwargs["n"],
"details": kwargs["n"] > 1,
# "max_new_tokens": kwargs.get('max_tokens', kwargs.get('max_new_tokens', 75)),
# "stop": ["\n", "\n\n"],
**kwargs,
},
}
payload["parameters"] = openai_to_hf(**payload["parameters"])
payload["parameters"]["temperature"] = max(
0.1, payload["parameters"]["temperature"],
)
# print(payload['parameters'])
# response = requests.post(self.url + "/generate", json=payload, headers=self.headers)
response = send_hftgi_request_v01_wrapped(
f"{self.url}:{random.Random().choice(self.ports)}" + "/generate",
url=self.url,
ports=tuple(self.ports),
json=payload,
headers=self.headers,
**self.http_request_kwargs,
)
try:
json_response = response.json()
# completions = json_response["generated_text"]
completions = [json_response["generated_text"]]
if (
"details" in json_response
and "best_of_sequences" in json_response["details"]
):
completions += [
x["generated_text"]
for x in json_response["details"]["best_of_sequences"]
]
response = {"prompt": prompt, "choices": [{"text": c} for c in completions]}
return response
except Exception:
print("Failed to parse JSON response:", response.text)
raise Exception("Received invalid JSON response from server")
@CacheMemory.cache(ignore=['arg'])
def send_hftgi_request_v01(arg, url, ports, **kwargs):
return requests.post(arg, **kwargs)
# @functools.lru_cache(maxsize=None if cache_turn_on else 0)
@NotebookCacheMemory.cache(ignore=['arg'])
def send_hftgi_request_v01_wrapped(arg, url, ports, **kwargs):
return send_hftgi_request_v01(arg, url, ports, **kwargs)
@CacheMemory.cache
def send_hftgi_request_v00(arg, **kwargs):
return requests.post(arg, **kwargs)
class HFClientVLLM(HFModel):
def __init__(self, model, port, model_type: Literal['chat', 'text'] = 'text', url="http:https://localhost", **kwargs):
super().__init__(model=model, is_client=True)
if isinstance(url, list):
self.urls = url
elif isinstance(url, str):
self.urls = [f'{url}:{port}']
else:
raise ValueError(f"The url provided to `HFClientVLLM` is neither a string nor a list of strings. It is of type {type(url)}.")
self.urls_const = tuple(self.urls)
self.port = port
self.model_type = model_type
self.headers = {"Content-Type": "application/json"}
self.kwargs |= kwargs
# kwargs needs to have model, port and url for the lm.copy() to work properly
self.kwargs.update({
'port': port,
'url': self.urls_const,
})
def _generate(self, prompt, **kwargs):
kwargs = {**self.kwargs, **kwargs}
# Round robin the urls.
url = self.urls.pop(0)
self.urls.append(url)
if self.model_type == "chat":
system_prompt = kwargs.get("system_prompt",None)
messages = [{"role": "user", "content": prompt}]
if system_prompt:
messages.insert(0, {"role": "system", "content": system_prompt})
payload = {
"model": self.kwargs["model"],
"messages": messages,
**kwargs,
}
response = send_hfvllm_request_v01_wrapped(
f"{url}/v1/chat/completions",
url=self.urls_const,
port=self.port,
json=payload,
headers=self.headers,
)
try:
json_response = response.json()
completions = json_response["choices"]
response = {
"prompt": prompt,
"choices": [{"text": c["message"]['content']} for c in completions],
}
return response
except Exception:
print("Failed to parse JSON response:", response.text)
raise Exception("Received invalid JSON response from server")
else:
payload = {
"model": self.kwargs["model"],
"prompt": prompt,
**kwargs,
}
response = send_hfvllm_request_v01_wrapped(
f"{url}/v1/completions",
url=self.urls_const,
port=self.port,
json=payload,
headers=self.headers,
)
try:
json_response = response.json()
completions = json_response["choices"]
response = {
"prompt": prompt,
"choices": [{"text": c["text"]} for c in completions],
}
return response
except Exception:
print("Failed to parse JSON response:", response.text)
raise Exception("Received invalid JSON response from server")
@CacheMemory.cache(ignore=['arg'])
def send_hfvllm_request_v01(arg, url, port, **kwargs):
return requests.post(arg, **kwargs)
# @functools.lru_cache(maxsize=None if cache_turn_on else 0)
@NotebookCacheMemory.cache(ignore=['arg'])
def send_hfvllm_request_v01_wrapped(arg, url, port, **kwargs):
return send_hftgi_request_v01(arg, url, port, **kwargs)
@CacheMemory.cache
def send_hfvllm_request_v00(arg, **kwargs):
return requests.post(arg, **kwargs)
@CacheMemory.cache
def send_hfvllm_chat_request_v00(arg, **kwargs):
return requests.post(arg, **kwargs)
class HFServerTGI:
def __init__(self, user_dir):
self.model_weights_dir = os.path.abspath(os.path.join(os.getcwd(), "text-generation-inference", user_dir))
if not os.path.exists(self.model_weights_dir):
os.makedirs(self.model_weights_dir)
def close_server(self, port):
process = subprocess.Popen(['docker', 'ps'], stdout=subprocess.PIPE, stderr=subprocess.PIPE)
stdout, _ = process.communicate()
print(stdout)
if stdout:
container_ids = stdout.decode().strip().split('\n')
container_ids = container_ids[1:]
for container_id in container_ids:
match = re.search(r'^([a-zA-Z0-9]+)', container_id)
if match:
container_id = match.group(1)
port_mapping = subprocess.check_output(['docker', 'port', container_id]).decode().strip()
if f'0.0.0.0:{port}' in port_mapping:
subprocess.run(['docker', 'stop', container_id], check=False)
def run_server(self, port, model_name=None, model_path=None, env_variable=None, gpus="all", num_shard=1, max_input_length=4000, max_total_tokens=4096, max_best_of=100):
self.close_server(port)
if model_path:
model_file_name = os.path.basename(model_path)
link_path = os.path.join(self.model_weights_dir, model_file_name)
shutil.copytree(model_path, link_path)
model_name = os.path.sep + os.path.basename(self.model_weights_dir) + os.path.sep + os.path.basename(model_path)
docker_command = f'docker run --gpus {gpus} --shm-size 1g -p {port}:80 -v {self.model_weights_dir}:{os.path.sep + os.path.basename(self.model_weights_dir)} -e {env_variable} ghcr.io/huggingface/text-generation-inference:1.1.0 --model-id {model_name} --num-shard {num_shard} --max-input-length {max_input_length} --max-total-tokens {max_total_tokens} --max-best-of {max_best_of}'
print(f"Connect Command: {docker_command}")
docker_process = subprocess.Popen(docker_command, shell=True, stdout=subprocess.PIPE, stderr=subprocess.STDOUT, text=True)
connected = False
output = []
while True:
line = docker_process.stdout.readline()
if not line:
break
output.append(line.strip())
if 'Connected' in line:
connected = True
break
if not connected:
print("Could not connect to server. Error log:")
for line in output:
print(line)
docker_process.terminate()
docker_process.wait()
class Together(HFModel):
def __init__(self, model, **kwargs):
super().__init__(model=model, is_client=True)
self.session = requests.Session()
self.api_base = os.getenv("TOGETHER_API_BASE")
self.token = os.getenv("TOGETHER_API_KEY")
self.model = model
self.use_inst_template = False
if any(keyword in self.model.lower() for keyword in ["inst", "instruct"]):
self.use_inst_template = True
stop_default = "\n\n---"
self.kwargs = {
"temperature": 0.0,
"max_tokens": 512,
"top_p": 1,
"top_k": 20,
"repetition_penalty": 1,
"n": 1,
"stop": stop_default if "stop" not in kwargs else kwargs["stop"],
**kwargs,
}
@backoff.on_exception(
backoff.expo,
ERRORS,
max_time=1000,
on_backoff=backoff_hdlr,
)
def _generate(self, prompt, use_chat_api=False, **kwargs):
url = f"{self.api_base}"
kwargs = {**self.kwargs, **kwargs}
stop = kwargs.get("stop")
temperature = kwargs.get("temperature")
max_tokens = kwargs.get("max_tokens", 150)
top_p = kwargs.get("top_p", 0.7)
top_k = kwargs.get("top_k", 50)
repetition_penalty = kwargs.get("repetition_penalty", 1)
prompt = f"[INST]{prompt}[/INST]" if self.use_inst_template else prompt
if use_chat_api:
url = f"{self.api_base}/chat/completions"
messages = [
{"role": "system", "content": "You are a helpful assistant. You must continue the user text directly without *any* additional interjections."},
{"role": "user", "content": prompt},
]
body = {
"model": self.model,
"messages": messages,
"temperature": temperature,
"max_tokens": max_tokens,
"top_p": top_p,
"top_k": top_k,
"repetition_penalty": repetition_penalty,
"stop": stop,
}
else:
body = {
"model": self.model,
"prompt": prompt,
"temperature": temperature,
"max_tokens": max_tokens,
"top_p": top_p,
"top_k": top_k,
"repetition_penalty": repetition_penalty,
"stop": stop,
}
headers = {"Authorization": f"Bearer {self.token}"}
try:
with self.session.post(url, headers=headers, json=body) as resp:
resp_json = resp.json()
if use_chat_api:
completions = [resp_json['output'].get('choices', [])[0].get('message', {}).get('content', "")]
else:
completions = [resp_json['output'].get('choices', [])[0].get('text', "")]
response = {"prompt": prompt, "choices": [{"text": c} for c in completions]}
return response
except Exception as e:
if resp_json:
print(f"resp_json:{resp_json}")
print(f"Failed to parse JSON response: {e}")
raise Exception("Received invalid JSON response from server")
class Anyscale(HFModel):
def __init__(self, model, **kwargs):
super().__init__(model=model, is_client=True)
self.session = requests.Session()
self.api_base = os.getenv("ANYSCALE_API_BASE")
self.token = os.getenv("ANYSCALE_API_KEY")
self.model = model
self.kwargs = {
"temperature": 0.0,
"n": 1,
**kwargs,
}
def _generate(self, prompt, use_chat_api=False, **kwargs):
url = f"{self.api_base}/completions"
kwargs = {**self.kwargs, **kwargs}
temperature = kwargs.get("temperature")
max_tokens = kwargs.get("max_tokens", 150)
if use_chat_api:
url = f"{self.api_base}/chat/completions"
messages = [
{"role": "system", "content": "You are a helpful assistant. You must continue the user text directly without *any* additional interjections."},
{"role": "user", "content": prompt},
]
body = {
"model": self.model,
"messages": messages,
"temperature": temperature,
"max_tokens": max_tokens,
}
else:
body = {
"model": self.model,
"prompt": f"[INST]{prompt}[/INST]",
"temperature": temperature,
"max_tokens": max_tokens,
}
headers = {"Authorization": f"Bearer {self.token}"}
try:
completions = []
for i in range(kwargs.get('n', 1)):
with self.session.post(url, headers=headers, json=body) as resp:
resp_json = resp.json()
if use_chat_api:
completions.extend([resp_json.get('choices', [])[0].get('message', {}).get('content', "")])
else:
completions.extend([resp_json.get('choices', [])[0].get('text', "")])
response = {"prompt": prompt, "choices": [{"text": c} for c in completions]}
return response
except Exception as e:
print(f"Failed to parse JSON response: {e}")
raise Exception("Received invalid JSON response from server")
class ChatModuleClient(HFModel):
def __init__(self, model, model_path):
super().__init__(model=model, is_client=True)
from mlc_chat import ChatConfig, ChatModule
self.cm = ChatModule(
model=model, lib_path=model_path, chat_config=ChatConfig(conv_template="LM"),
)
def _generate(self, prompt, **kwargs):
output = self.cm.generate(
prompt=prompt,
)
try:
completions = [{"text": output}]
response = {"prompt": prompt, "choices": completions}
return response
except Exception:
print("Failed to parse output:", response.text)
raise Exception("Received invalid output")
class HFClientSGLang(HFModel):
def __init__(self, model, port, url="http:https://localhost", **kwargs):
super().__init__(model=model, is_client=True)
self.url = f"{url}:{port}"
self.headers = {"Content-Type": "application/json"}
self.kwargs = {
"temperature": 0.01,
"max_tokens": 75,
"top_p": 0.97,
"n": 1,
"stop": ["\n", "\n\n"],
**kwargs,
}
def _generate(self, prompt, **kwargs):
kwargs = {**self.kwargs, **kwargs}
payload = {
"model": kwargs.get("model", "default"),
"prompt": prompt,
**kwargs,
}
response = send_hfsglang_request_v00(
f"{self.url}/v1/completions",
json=payload,
headers=self.headers,
)
try:
json_response = response.json()
completions = json_response["choices"]
response = {
"prompt": prompt,
"choices": [{"text": c["text"]} for c in completions],
}
return response
except Exception:
print("Failed to parse JSON response:", response.text)
raise Exception("Received invalid JSON response from server")
@CacheMemory.cache
def send_hfsglang_request_v00(arg, **kwargs):
return requests.post(arg, **kwargs)