Spaces:
Runtime error
Runtime error
:gem: [Feature] Moduralize TokenChecker, and fix gated model repos with alternatives
Browse files- messagers/token_checker.py +44 -0
- networks/huggingchat_streamer.py +4 -48
- networks/huggingface_streamer.py +5 -29
messagers/token_checker.py
ADDED
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from tclogger import logger
|
2 |
+
from transformers import AutoTokenizer
|
3 |
+
|
4 |
+
from constants.models import MODEL_MAP, TOKEN_LIMIT_MAP, TOKEN_RESERVED
|
5 |
+
|
6 |
+
|
7 |
+
class TokenChecker:
|
8 |
+
def __init__(self, input_str: str, model: str):
|
9 |
+
self.input_str = input_str
|
10 |
+
|
11 |
+
if model in MODEL_MAP.keys():
|
12 |
+
self.model = model
|
13 |
+
else:
|
14 |
+
self.model = "mixtral-8x7b"
|
15 |
+
|
16 |
+
self.model_fullname = MODEL_MAP[self.model]
|
17 |
+
|
18 |
+
# As some models are gated, we need to fetch tokenizers from alternatives
|
19 |
+
GATED_MODEL_MAP = {
|
20 |
+
"llama3-70b": "NousResearch/Meta-Llama-3-70B",
|
21 |
+
"gemma-7b": "unsloth/gemma-7b",
|
22 |
+
"mistral-7b": "dfurman/Mistral-7B-Instruct-v0.2",
|
23 |
+
"mixtral-8x7b": "dfurman/Mixtral-8x7B-Instruct-v0.1",
|
24 |
+
}
|
25 |
+
if self.model in GATED_MODEL_MAP.keys():
|
26 |
+
self.tokenizer = AutoTokenizer.from_pretrained(GATED_MODEL_MAP[self.model])
|
27 |
+
else:
|
28 |
+
self.tokenizer = AutoTokenizer.from_pretrained(self.model_fullname)
|
29 |
+
|
30 |
+
def count_tokens(self):
|
31 |
+
token_count = len(self.tokenizer.encode(self.input_str))
|
32 |
+
logger.note(f"Prompt Token Count: {token_count}")
|
33 |
+
return token_count
|
34 |
+
|
35 |
+
def get_token_limit(self):
|
36 |
+
return TOKEN_LIMIT_MAP[self.model]
|
37 |
+
|
38 |
+
def get_token_redundancy(self):
|
39 |
+
return int(self.get_token_limit() - TOKEN_RESERVED - self.count_tokens())
|
40 |
+
|
41 |
+
def check_token_limit(self):
|
42 |
+
if self.get_token_redundancy() <= 0:
|
43 |
+
raise ValueError(f"Prompt exceeded token limit: {self.get_token_limit()}")
|
44 |
+
return True
|
networks/huggingchat_streamer.py
CHANGED
@@ -2,59 +2,15 @@ import copy
|
|
2 |
import json
|
3 |
import re
|
4 |
import requests
|
5 |
-
import uuid
|
6 |
|
7 |
-
# from curl_cffi import requests
|
8 |
from tclogger import logger
|
9 |
-
|
10 |
-
|
11 |
-
from constants.models import (
|
12 |
-
MODEL_MAP,
|
13 |
-
STOP_SEQUENCES_MAP,
|
14 |
-
TOKEN_LIMIT_MAP,
|
15 |
-
TOKEN_RESERVED,
|
16 |
-
)
|
17 |
from constants.envs import PROXIES
|
18 |
-
from constants.headers import
|
19 |
-
REQUESTS_HEADERS,
|
20 |
-
HUGGINGCHAT_POST_HEADERS,
|
21 |
-
HUGGINGCHAT_SETTINGS_POST_DATA,
|
22 |
-
)
|
23 |
from messagers.message_outputer import OpenaiStreamOutputer
|
24 |
from messagers.message_composer import MessageComposer
|
25 |
-
|
26 |
-
|
27 |
-
class TokenChecker:
|
28 |
-
def __init__(self, input_str: str, model: str):
|
29 |
-
self.input_str = input_str
|
30 |
-
|
31 |
-
if model in MODEL_MAP.keys():
|
32 |
-
self.model = model
|
33 |
-
else:
|
34 |
-
self.model = "mixtral-8x7b"
|
35 |
-
|
36 |
-
self.model_fullname = MODEL_MAP[self.model]
|
37 |
-
|
38 |
-
if self.model == "llama3-70b":
|
39 |
-
# As original llama3 repo is gated and requires auth,
|
40 |
-
# I use NousResearch's version as a workaround
|
41 |
-
self.tokenizer = AutoTokenizer.from_pretrained(
|
42 |
-
"NousResearch/Meta-Llama-3-70B"
|
43 |
-
)
|
44 |
-
else:
|
45 |
-
self.tokenizer = AutoTokenizer.from_pretrained(self.model_fullname)
|
46 |
-
|
47 |
-
def count_tokens(self):
|
48 |
-
token_count = len(self.tokenizer.encode(self.input_str))
|
49 |
-
logger.note(f"Prompt Token Count: {token_count}")
|
50 |
-
return token_count
|
51 |
-
|
52 |
-
def check_token_limit(self):
|
53 |
-
token_limit = TOKEN_LIMIT_MAP[self.model]
|
54 |
-
token_redundancy = int(token_limit - TOKEN_RESERVED - self.count_tokens())
|
55 |
-
if token_redundancy <= 0:
|
56 |
-
raise ValueError(f"Prompt exceeded token limit: {token_limit}")
|
57 |
-
return True
|
58 |
|
59 |
|
60 |
class HuggingchatRequester:
|
|
|
2 |
import json
|
3 |
import re
|
4 |
import requests
|
|
|
5 |
|
|
|
6 |
from tclogger import logger
|
7 |
+
|
8 |
+
from constants.models import MODEL_MAP
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
from constants.envs import PROXIES
|
10 |
+
from constants.headers import HUGGINGCHAT_POST_HEADERS, HUGGINGCHAT_SETTINGS_POST_DATA
|
|
|
|
|
|
|
|
|
11 |
from messagers.message_outputer import OpenaiStreamOutputer
|
12 |
from messagers.message_composer import MessageComposer
|
13 |
+
from messagers.token_checker import TokenChecker
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
|
16 |
class HuggingchatRequester:
|
networks/huggingface_streamer.py
CHANGED
@@ -2,18 +2,11 @@ import json
|
|
2 |
import re
|
3 |
import requests
|
4 |
|
5 |
-
|
6 |
from tclogger import logger
|
7 |
-
from
|
8 |
-
|
9 |
-
from constants.models import (
|
10 |
-
MODEL_MAP,
|
11 |
-
STOP_SEQUENCES_MAP,
|
12 |
-
TOKEN_LIMIT_MAP,
|
13 |
-
TOKEN_RESERVED,
|
14 |
-
)
|
15 |
from constants.envs import PROXIES
|
16 |
from messagers.message_outputer import OpenaiStreamOutputer
|
|
|
17 |
|
18 |
|
19 |
class HuggingfaceStreamer:
|
@@ -25,13 +18,6 @@ class HuggingfaceStreamer:
|
|
25 |
self.model_fullname = MODEL_MAP[self.model]
|
26 |
self.message_outputer = OpenaiStreamOutputer(model=self.model)
|
27 |
|
28 |
-
if self.model == "gemma-7b":
|
29 |
-
# this is not wrong, as repo `google/gemma-7b-it` is gated and must authenticate to access it
|
30 |
-
# so I use mistral-7b as a fallback
|
31 |
-
self.tokenizer = AutoTokenizer.from_pretrained(MODEL_MAP["mistral-7b"])
|
32 |
-
else:
|
33 |
-
self.tokenizer = AutoTokenizer.from_pretrained(self.model_fullname)
|
34 |
-
|
35 |
def parse_line(self, line):
|
36 |
line = line.decode("utf-8")
|
37 |
line = re.sub(r"data:\s*", "", line)
|
@@ -42,12 +28,6 @@ class HuggingfaceStreamer:
|
|
42 |
logger.err(data)
|
43 |
return content
|
44 |
|
45 |
-
def count_tokens(self, text):
|
46 |
-
tokens = self.tokenizer.encode(text)
|
47 |
-
token_count = len(tokens)
|
48 |
-
logger.note(f"Prompt Token Count: {token_count}")
|
49 |
-
return token_count
|
50 |
-
|
51 |
def chat_response(
|
52 |
self,
|
53 |
prompt: str = None,
|
@@ -80,16 +60,12 @@ class HuggingfaceStreamer:
|
|
80 |
top_p = max(top_p, 0.01)
|
81 |
top_p = min(top_p, 0.99)
|
82 |
|
83 |
-
|
84 |
-
TOKEN_LIMIT_MAP[self.model] - TOKEN_RESERVED - self.count_tokens(prompt)
|
85 |
-
)
|
86 |
-
if token_limit <= 0:
|
87 |
-
raise ValueError("Prompt exceeded token limit!")
|
88 |
|
89 |
if max_new_tokens is None or max_new_tokens <= 0:
|
90 |
-
max_new_tokens =
|
91 |
else:
|
92 |
-
max_new_tokens = min(max_new_tokens,
|
93 |
|
94 |
# References:
|
95 |
# huggingface_hub/inference/_client.py:
|
|
|
2 |
import re
|
3 |
import requests
|
4 |
|
|
|
5 |
from tclogger import logger
|
6 |
+
from constants.models import MODEL_MAP, STOP_SEQUENCES_MAP
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
from constants.envs import PROXIES
|
8 |
from messagers.message_outputer import OpenaiStreamOutputer
|
9 |
+
from messagers.token_checker import TokenChecker
|
10 |
|
11 |
|
12 |
class HuggingfaceStreamer:
|
|
|
18 |
self.model_fullname = MODEL_MAP[self.model]
|
19 |
self.message_outputer = OpenaiStreamOutputer(model=self.model)
|
20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
def parse_line(self, line):
|
22 |
line = line.decode("utf-8")
|
23 |
line = re.sub(r"data:\s*", "", line)
|
|
|
28 |
logger.err(data)
|
29 |
return content
|
30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
def chat_response(
|
32 |
self,
|
33 |
prompt: str = None,
|
|
|
60 |
top_p = max(top_p, 0.01)
|
61 |
top_p = min(top_p, 0.99)
|
62 |
|
63 |
+
checker = TokenChecker(input_str=prompt, model=self.model)
|
|
|
|
|
|
|
|
|
64 |
|
65 |
if max_new_tokens is None or max_new_tokens <= 0:
|
66 |
+
max_new_tokens = checker.get_token_redundancy()
|
67 |
else:
|
68 |
+
max_new_tokens = min(max_new_tokens, checker.get_token_redundancy())
|
69 |
|
70 |
# References:
|
71 |
# huggingface_hub/inference/_client.py:
|