Spaces:
Running
on
Zero
Running
on
Zero
Upload 2 files
Browse files- formatter.py +43 -0
- llmdolphin.py +4 -1
formatter.py
ADDED
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from llama_cpp_agent.messages_formatter import MessagesFormatter, PromptMarkers, Roles
|
2 |
+
|
3 |
+
mistral_v1_markers = {
|
4 |
+
Roles.system: PromptMarkers(""" [INST]""", """ [/INST] Understood.</s>"""),
|
5 |
+
Roles.user: PromptMarkers(""" [INST]""", """ [/INST]"""),
|
6 |
+
Roles.assistant: PromptMarkers(" ", "</s>"),
|
7 |
+
Roles.tool: PromptMarkers("", ""),
|
8 |
+
}
|
9 |
+
|
10 |
+
mistral_v1_formatter = MessagesFormatter(
|
11 |
+
pre_prompt="",
|
12 |
+
prompt_markers=mistral_v1_markers,
|
13 |
+
include_sys_prompt_in_first_user_message=False,
|
14 |
+
default_stop_sequences=["</s>"]
|
15 |
+
)
|
16 |
+
|
17 |
+
mistral_v2_markers = {
|
18 |
+
Roles.system: PromptMarkers("""[INST] """, """[/INST] Understood.</s>"""),
|
19 |
+
Roles.user: PromptMarkers("""[INST] """, """[/INST]"""),
|
20 |
+
Roles.assistant: PromptMarkers(" ", "</s>"),
|
21 |
+
Roles.tool: PromptMarkers("", ""),
|
22 |
+
}
|
23 |
+
|
24 |
+
mistral_v2_formatter = MessagesFormatter(
|
25 |
+
pre_prompt="",
|
26 |
+
prompt_markers=mistral_v2_markers,
|
27 |
+
include_sys_prompt_in_first_user_message=False,
|
28 |
+
default_stop_sequences=["</s>"]
|
29 |
+
)
|
30 |
+
|
31 |
+
mistral_v3_tekken_markers = {
|
32 |
+
Roles.system: PromptMarkers("""[INST]""", """[/INST]Understood.</s>"""),
|
33 |
+
Roles.user: PromptMarkers("""[INST]""", """[/INST]"""),
|
34 |
+
Roles.assistant: PromptMarkers("", "</s>"),
|
35 |
+
Roles.tool: PromptMarkers("", ""),
|
36 |
+
}
|
37 |
+
|
38 |
+
mistral_v3_tekken_formatter = MessagesFormatter(
|
39 |
+
pre_prompt="",
|
40 |
+
prompt_markers=mistral_v3_tekken_markers,
|
41 |
+
include_sys_prompt_in_first_user_message=False,
|
42 |
+
default_stop_sequences=["</s>"]
|
43 |
+
)
|
llmdolphin.py
CHANGED
@@ -8,7 +8,7 @@ from llama_cpp_agent.chat_history.messages import Roles
|
|
8 |
from ja_to_danbooru.ja_to_danbooru import jatags_to_danbooru_tags
|
9 |
import wrapt_timeout_decorator
|
10 |
from pathlib import Path
|
11 |
-
|
12 |
|
13 |
llm_models_dir = "./llm_models"
|
14 |
llm_models = {
|
@@ -821,6 +821,9 @@ llm_formats = {
|
|
821 |
"DeepSeek Coder v2": MessagesFormatterType.DEEP_SEEK_CODER_2,
|
822 |
"Gemma 2": MessagesFormatterType.ALPACA,
|
823 |
"Qwen2": MessagesFormatterType.OPEN_CHAT,
|
|
|
|
|
|
|
824 |
}
|
825 |
# https://github.com/Maximilian-Winter/llama-cpp-agent
|
826 |
llm_languages = ["English", "Japanese", "Chinese", "Korean", "Spanish", "Portuguese", "German", "French", "Finnish", "Russian"]
|
|
|
8 |
from ja_to_danbooru.ja_to_danbooru import jatags_to_danbooru_tags
|
9 |
import wrapt_timeout_decorator
|
10 |
from pathlib import Path
|
11 |
+
from formatter import mistral_v1_formatter, mistral_v2_formatter, mistral_v3_tekken_formatter
|
12 |
|
13 |
llm_models_dir = "./llm_models"
|
14 |
llm_models = {
|
|
|
821 |
"DeepSeek Coder v2": MessagesFormatterType.DEEP_SEEK_CODER_2,
|
822 |
"Gemma 2": MessagesFormatterType.ALPACA,
|
823 |
"Qwen2": MessagesFormatterType.OPEN_CHAT,
|
824 |
+
"Mistral Tokenizer V1": mistral_v1_formatter,
|
825 |
+
"Mistral Tokenizer V2": mistral_v2_formatter,
|
826 |
+
"Mistral Tokenizer V3 - Tekken": mistral_v3_tekken_formatter,
|
827 |
}
|
828 |
# https://github.com/Maximilian-Winter/llama-cpp-agent
|
829 |
llm_languages = ["English", "Japanese", "Chinese", "Korean", "Spanish", "Portuguese", "German", "French", "Finnish", "Russian"]
|