Paul Rock
commited on
Commit
•
a33d530
1
Parent(s):
aa68cf9
Initial commit
Browse files- .gitignore +1 -0
- README.md +9 -0
- adapter_config.json +28 -0
- adapter_model.bin +3 -0
- chat.json +10 -0
- config.json +36 -0
- generation_config.json +13 -0
- special_tokens_map.json +37 -0
- test_lora.py +141 -0
- tokenizer.model +3 -0
- tokenizer_config.json +43 -0
.gitignore
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
/.idea/
|
README.md
CHANGED
@@ -1,3 +1,12 @@
|
|
1 |
---
|
2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
1 |
---
|
2 |
license: mit
|
3 |
+
datasets:
|
4 |
+
- d0rj/gsm8k-ru
|
5 |
+
- d0rj/alpaca-cleaned-ru
|
6 |
+
- IlyaGusev/ru_turbo_alpaca
|
7 |
+
- IlyaGusev/ru_turbo_alpaca_evol_instruct
|
8 |
+
language:
|
9 |
+
- ru
|
10 |
+
library_name: peft
|
11 |
+
pipeline_tag: text-generation
|
12 |
---
|
adapter_config.json
ADDED
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"alpha_pattern": {},
|
3 |
+
"auto_mapping": null,
|
4 |
+
"base_model_name_or_path": "/home/pasha/Documents/Repository/gpt/saiga-custom/Yarn-Mistral-7b-128k",
|
5 |
+
"bias": "none",
|
6 |
+
"fan_in_fan_out": false,
|
7 |
+
"inference_mode": true,
|
8 |
+
"init_lora_weights": true,
|
9 |
+
"layers_pattern": null,
|
10 |
+
"layers_to_transform": null,
|
11 |
+
"loftq_config": {},
|
12 |
+
"lora_alpha": 16,
|
13 |
+
"lora_dropout": 0.05,
|
14 |
+
"megatron_config": null,
|
15 |
+
"megatron_core": "megatron.core",
|
16 |
+
"modules_to_save": null,
|
17 |
+
"peft_type": "LORA",
|
18 |
+
"r": 16,
|
19 |
+
"rank_pattern": {},
|
20 |
+
"revision": null,
|
21 |
+
"target_modules": [
|
22 |
+
"q_proj",
|
23 |
+
"o_proj",
|
24 |
+
"k_proj",
|
25 |
+
"v_proj"
|
26 |
+
],
|
27 |
+
"task_type": "CAUSAL_LM"
|
28 |
+
}
|
adapter_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:972e34d18e05e346f738031a80b365cb82c29c0b9d6e0674ee4899ff5541d2fe
|
3 |
+
size 54618762
|
chat.json
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"system_prompt": "Ты — PavelGPT, русскоязычный автоматический ассистент. Ты разговариваешь с людьми и помогаешь им.",
|
3 |
+
"system_message_template": "<s>{role}\n{content}</s>\n",
|
4 |
+
"user_message_template": "<s>{role}\n{content}</s>\n",
|
5 |
+
"bot_message_template": "<s>{role}\n{content}</s>\n",
|
6 |
+
"user_role": "user",
|
7 |
+
"bot_role": "bot",
|
8 |
+
"system_role": "system",
|
9 |
+
"suffix": ""
|
10 |
+
}
|
config.json
ADDED
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "NousResearch/Yarn-Mistral-7b-128k",
|
3 |
+
"architectures": [
|
4 |
+
"MistralForCausalLM"
|
5 |
+
],
|
6 |
+
"auto_map": {
|
7 |
+
"AutoConfig": "configuration_mistral.MistralConfig",
|
8 |
+
"AutoModelForCausalLM": "modeling_mistral_yarn.MistralForCausalLM"
|
9 |
+
},
|
10 |
+
"bos_token_id": 1,
|
11 |
+
"eos_token_id": 2,
|
12 |
+
"hidden_act": "silu",
|
13 |
+
"hidden_size": 4096,
|
14 |
+
"initializer_range": 0.02,
|
15 |
+
"intermediate_size": 14336,
|
16 |
+
"max_position_embeddings": 32768,
|
17 |
+
"max_sequence_length": 131072,
|
18 |
+
"model_type": "mistral",
|
19 |
+
"num_attention_heads": 32,
|
20 |
+
"num_hidden_layers": 32,
|
21 |
+
"num_key_value_heads": 8,
|
22 |
+
"rms_norm_eps": 1e-05,
|
23 |
+
"rope_scaling": {
|
24 |
+
"factor": 16.0,
|
25 |
+
"finetuned": true,
|
26 |
+
"original_max_position_embeddings": 8192,
|
27 |
+
"type": "yarn"
|
28 |
+
},
|
29 |
+
"rope_theta": 10000.0,
|
30 |
+
"sliding_window": 131072,
|
31 |
+
"tie_word_embeddings": false,
|
32 |
+
"torch_dtype": "bfloat16",
|
33 |
+
"transformers_version": "4.35.0.dev0",
|
34 |
+
"use_cache": true,
|
35 |
+
"vocab_size": 32000
|
36 |
+
}
|
generation_config.json
ADDED
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"pad_token_id": 0,
|
3 |
+
"unk_token_id": 0,
|
4 |
+
"bos_token_id": 1,
|
5 |
+
"eos_token_id": 2,
|
6 |
+
"temperature": 0.2,
|
7 |
+
"top_p": 0.2,
|
8 |
+
"top_k": 20,
|
9 |
+
"do_sample": true,
|
10 |
+
"max_new_tokens": 4096,
|
11 |
+
"repetition_penalty": 1.2,
|
12 |
+
"no_repeat_ngram_size": 15
|
13 |
+
}
|
special_tokens_map.json
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": {
|
3 |
+
"content": "<s>",
|
4 |
+
"lstrip": false,
|
5 |
+
"normalized": false,
|
6 |
+
"rstrip": false,
|
7 |
+
"single_word": false
|
8 |
+
},
|
9 |
+
"eos_token": {
|
10 |
+
"content": "</s>",
|
11 |
+
"lstrip": false,
|
12 |
+
"normalized": false,
|
13 |
+
"rstrip": false,
|
14 |
+
"single_word": false
|
15 |
+
},
|
16 |
+
"pad_token": {
|
17 |
+
"content": "<unk>",
|
18 |
+
"lstrip": false,
|
19 |
+
"normalized": false,
|
20 |
+
"rstrip": false,
|
21 |
+
"single_word": false
|
22 |
+
},
|
23 |
+
"sep_token": {
|
24 |
+
"content": "<s>",
|
25 |
+
"lstrip": false,
|
26 |
+
"normalized": false,
|
27 |
+
"rstrip": false,
|
28 |
+
"single_word": false
|
29 |
+
},
|
30 |
+
"unk_token": {
|
31 |
+
"content": "<unk>",
|
32 |
+
"lstrip": false,
|
33 |
+
"normalized": false,
|
34 |
+
"rstrip": false,
|
35 |
+
"single_word": false
|
36 |
+
}
|
37 |
+
}
|
test_lora.py
ADDED
@@ -0,0 +1,141 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import logging
|
3 |
+
from peft import PeftModel, PeftConfig
|
4 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
|
5 |
+
|
6 |
+
# MODEL_NAME = "IlyaGusev/gigasaiga_lora"
|
7 |
+
# MODEL_NAME = "evilfreelancer/ruGPT-3.5-13B-lora"
|
8 |
+
# MODEL_NAME = "evilfreelancer/saiga_mistral_7b_128k_lora"
|
9 |
+
# MODEL_NAME = "./yarn_mistral_7b_128k"
|
10 |
+
MODEL_NAME = "./yarn_mistral_7b_128k_yakovlev"
|
11 |
+
|
12 |
+
DEFAULT_MESSAGE_TEMPLATE = "<s>{role}\n{content}</s>\n"
|
13 |
+
# DEFAULT_SYSTEM_PROMPT = """
|
14 |
+
# Ты — Saiga 2, русскоязычный автоматический ассистент. Ты разговариваешь с людьми и помогаешь им.
|
15 |
+
# """
|
16 |
+
DEFAULT_SYSTEM_PROMPT = """
|
17 |
+
Ты - Иван Яковлевич Яковлев. Ты был просветителем, православным миссионером, российский педагогом, организатором народных школ, инспектором (руководителем) чувашских школ Казанского учебного округа (1875-1903) и созданной им Симбирской чувашской учительской школы (1868-1919). Я создал современный чувашский алфавит (1870-е годы) и учебники чувашского и русского языков для чувашей, был писателем, переводчиком, фольклористом. Действительный статский советник (1904).
|
18 |
+
"""
|
19 |
+
|
20 |
+
|
21 |
+
class Conversation:
|
22 |
+
def __init__(
|
23 |
+
self,
|
24 |
+
message_template=DEFAULT_MESSAGE_TEMPLATE,
|
25 |
+
system_prompt=DEFAULT_SYSTEM_PROMPT,
|
26 |
+
start_token_id=2,
|
27 |
+
# Bot token may be a list or single int
|
28 |
+
bot_token_id=10093, # yarn_mistral_7b_128k
|
29 |
+
# bot_token_id=46787, # rugpt35_13b
|
30 |
+
# int (amount of questions and answers) or None (unlimited)
|
31 |
+
history_limit=None,
|
32 |
+
):
|
33 |
+
self.logger = logging.getLogger('Conversation')
|
34 |
+
self.message_template = message_template
|
35 |
+
self.start_token_id = start_token_id
|
36 |
+
self.bot_token_id = bot_token_id
|
37 |
+
self.history_limit = history_limit
|
38 |
+
self.messages = [{
|
39 |
+
"role": "system",
|
40 |
+
"content": system_prompt
|
41 |
+
}]
|
42 |
+
|
43 |
+
def get_start_token_id(self):
|
44 |
+
return self.start_token_id
|
45 |
+
|
46 |
+
def get_bot_token_id(self):
|
47 |
+
return self.bot_token_id
|
48 |
+
|
49 |
+
def add_message(self, role, message):
|
50 |
+
self.messages.append({
|
51 |
+
"role": role,
|
52 |
+
"content": message
|
53 |
+
})
|
54 |
+
self.trim_history()
|
55 |
+
|
56 |
+
def add_user_message(self, message):
|
57 |
+
self.add_message("user", message)
|
58 |
+
|
59 |
+
def add_bot_message(self, message):
|
60 |
+
self.add_message("assistant", message)
|
61 |
+
|
62 |
+
def trim_history(self):
|
63 |
+
if self.history_limit is not None and len(self.messages) > self.history_limit + 1:
|
64 |
+
overflow = len(self.messages) - (self.history_limit + 1)
|
65 |
+
self.messages = [self.messages[0]] + self.messages[overflow + 1:] # remove old messages except system
|
66 |
+
|
67 |
+
def get_prompt(self, tokenizer):
|
68 |
+
final_text = ""
|
69 |
+
# print(self.messages)
|
70 |
+
for message in self.messages:
|
71 |
+
message_text = self.message_template.format(**message)
|
72 |
+
final_text += message_text
|
73 |
+
|
74 |
+
# Bot token id may be an array
|
75 |
+
if isinstance(self.bot_token_id, (list, tuple)):
|
76 |
+
final_text += tokenizer.decode([self.start_token_id] + self.bot_token_id)
|
77 |
+
else:
|
78 |
+
final_text += tokenizer.decode([self.start_token_id, self.bot_token_id])
|
79 |
+
|
80 |
+
return final_text.strip()
|
81 |
+
|
82 |
+
|
83 |
+
def generate(model, tokenizer, prompt, generation_config):
|
84 |
+
data = tokenizer(prompt, return_tensors="pt")
|
85 |
+
data = {k: v.to(model.device) for k, v in data.items()}
|
86 |
+
output_ids = model.generate(
|
87 |
+
**data,
|
88 |
+
max_length=10240,
|
89 |
+
generation_config=generation_config
|
90 |
+
)[0]
|
91 |
+
output_ids = output_ids[len(data["input_ids"][0]):]
|
92 |
+
output = tokenizer.decode(output_ids, skip_special_tokens=True)
|
93 |
+
return output.strip()
|
94 |
+
|
95 |
+
|
96 |
+
config = PeftConfig.from_pretrained(MODEL_NAME)
|
97 |
+
model = AutoModelForCausalLM.from_pretrained(
|
98 |
+
config.base_model_name_or_path,
|
99 |
+
load_in_8bit=True,
|
100 |
+
torch_dtype=torch.float16,
|
101 |
+
device_map="auto",
|
102 |
+
use_flash_attention_2=True,
|
103 |
+
)
|
104 |
+
model = PeftModel.from_pretrained(
|
105 |
+
model,
|
106 |
+
MODEL_NAME,
|
107 |
+
torch_dtype=torch.float16
|
108 |
+
)
|
109 |
+
model.eval()
|
110 |
+
|
111 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=False)
|
112 |
+
generation_config = GenerationConfig.from_pretrained(MODEL_NAME)
|
113 |
+
print(generation_config)
|
114 |
+
|
115 |
+
conversation = Conversation()
|
116 |
+
while True:
|
117 |
+
user_message = input("User: ")
|
118 |
+
|
119 |
+
# Reset chat command
|
120 |
+
if user_message.strip() == "/reset":
|
121 |
+
conversation = Conversation()
|
122 |
+
print("History reset completed!")
|
123 |
+
continue
|
124 |
+
|
125 |
+
# Skip empty messages from user
|
126 |
+
if user_message.strip() == "":
|
127 |
+
continue
|
128 |
+
|
129 |
+
conversation.add_user_message(user_message)
|
130 |
+
prompt = conversation.get_prompt(tokenizer)
|
131 |
+
output = generate(
|
132 |
+
model=model,
|
133 |
+
tokenizer=tokenizer,
|
134 |
+
prompt=prompt,
|
135 |
+
generation_config=generation_config
|
136 |
+
)
|
137 |
+
conversation.add_bot_message(output)
|
138 |
+
print("Bot:", output)
|
139 |
+
print()
|
140 |
+
print("==============================")
|
141 |
+
print()
|
tokenizer.model
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
|
3 |
+
size 493443
|
tokenizer_config.json
ADDED
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"add_bos_token": false,
|
3 |
+
"add_eos_token": false,
|
4 |
+
"added_tokens_decoder": {
|
5 |
+
"0": {
|
6 |
+
"content": "<unk>",
|
7 |
+
"lstrip": false,
|
8 |
+
"normalized": false,
|
9 |
+
"rstrip": false,
|
10 |
+
"single_word": false,
|
11 |
+
"special": true
|
12 |
+
},
|
13 |
+
"1": {
|
14 |
+
"content": "<s>",
|
15 |
+
"lstrip": false,
|
16 |
+
"normalized": false,
|
17 |
+
"rstrip": false,
|
18 |
+
"single_word": false,
|
19 |
+
"special": true
|
20 |
+
},
|
21 |
+
"2": {
|
22 |
+
"content": "</s>",
|
23 |
+
"lstrip": false,
|
24 |
+
"normalized": false,
|
25 |
+
"rstrip": false,
|
26 |
+
"single_word": false,
|
27 |
+
"special": true
|
28 |
+
}
|
29 |
+
},
|
30 |
+
"bos_token": "<s>",
|
31 |
+
"clean_up_tokenization_spaces": false,
|
32 |
+
"eos_token": "</s>",
|
33 |
+
"legacy": true,
|
34 |
+
"model_max_length": 32768,
|
35 |
+
"pad_token": "<unk>",
|
36 |
+
"padding_side": "left",
|
37 |
+
"sep_token": "<s>",
|
38 |
+
"sp_model_kwargs": {},
|
39 |
+
"spaces_between_special_tokens": false,
|
40 |
+
"tokenizer_class": "LlamaTokenizer",
|
41 |
+
"unk_token": "<unk>",
|
42 |
+
"use_default_system_prompt": false
|
43 |
+
}
|