Saiga
Collection
LLaMA-based or Mistral-based Russian chat models
•
8 items
•
Updated
•
23
YAML Metadata
Warning:
The pipeline tag "conversational" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, text2text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, any-to-any, other
Saiga2 70B, Russian LLaMA2-based chatbot
Based on LLaMA-2 70B fp16
Llama.cpp version: link
This is an adapter-only version.
Training code: link
WARNING: Avoid using V100 (in Colab, for example). Outputs are much worse in this case.
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
MODEL_NAME = "IlyaGusev/saiga2_70b_lora"
DEFAULT_MESSAGE_TEMPLATE = "<s>{role}\n{content}</s>\n"
DEFAULT_SYSTEM_PROMPT = "Ты — Сайга, русскоязычный автоматический ассистент. Ты разговариваешь с людьми и помогаешь им."
class Conversation:
def __init__(
self,
message_template=DEFAULT_MESSAGE_TEMPLATE,
system_prompt=DEFAULT_SYSTEM_PROMPT,
start_token_id=1,
bot_token_id=9225
):
self.message_template = message_template
self.start_token_id = start_token_id
self.bot_token_id = bot_token_id
self.messages = [{
"role": "system",
"content": system_prompt
}]
def get_start_token_id(self):
return self.start_token_id
def get_bot_token_id(self):
return self.bot_token_id
def add_user_message(self, message):
self.messages.append({
"role": "user",
"content": message
})
def add_bot_message(self, message):
self.messages.append({
"role": "bot",
"content": message
})
def get_prompt(self, tokenizer):
final_text = ""
for message in self.messages:
message_text = self.message_template.format(**message)
final_text += message_text
final_text += tokenizer.decode([self.start_token_id, self.bot_token_id])
return final_text.strip()
def generate(model, tokenizer, prompt, generation_config):
data = tokenizer(prompt, return_tensors="pt")
data = {k: v.to(model.device) for k, v in data.items()}
output_ids = model.generate(
**data,
generation_config=generation_config
)[0]
output_ids = output_ids[len(data["input_ids"][0]):]
output = tokenizer.decode(output_ids, skip_special_tokens=True)
return output.strip()
config = PeftConfig.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(
config.base_model_name_or_path,
load_in_8bit=True,
torch_dtype=torch.float16,
device_map="auto"
)
model = PeftModel.from_pretrained(
model,
MODEL_NAME,
torch_dtype=torch.float16
)
model.eval()
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, use_fast=False)
generation_config = GenerationConfig.from_pretrained(MODEL_NAME)
print(generation_config)
inputs = ["Почему трава зеленая?", "Сочини длинный рассказ, обязательно упоминая следующие объекты. Дано: Таня, мяч"]
for inp in inputs:
conversation = Conversation()
conversation.add_user_message(inp)
prompt = conversation.get_prompt(tokenizer)
output = generate(model, tokenizer, prompt, generation_config)
print(inp)
print(output)
print()
print("==============================")
print()
Examples:
User: Почему трава зеленая?
Saiga:
User: Сочини длинный рассказ, обязательно упоминая следующие объекты. Дано: Таня, мяч
Saiga:
v1:
- dataset code revision 0dbd022613874fcda915f588f4a3292e137017d2
- wandb link
- 5 datasets: ru_turbo_alpaca, ru_sharegpt_cleaned, oasst1_ru_main_branch, gpt_roleplay_realm, ru_instruct_gpt4
- Datasets merging script: create_chat_set.py
- saiga2_70b vs gpt-3.5-turbo: 91-10-75