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, audio-text-to-text, 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
ruGPT-3.5 13B GGML
Welcome to the adapter-only version of ruGPT-3.5 13B GGML. This model is built upon the foundation of ruGPT-3.5-13B.
📌 Important: This model was trained using settings identical to GigaSaiga, but incorporates additional dataset.
🔗 Training code is here.
Code sample
from llm_rs import AutoModel, GenerationConfig as GConfig
from transformers import AutoTokenizer, GenerationConfig
MODEL_NAME = "evilfreelancer/ruGPT-3.5-13B-ggml"
DEFAULT_MESSAGE_TEMPLATE = "<s>{role}\n{content}</s>\n"
DEFAULT_SYSTEM_PROMPT = "Ты — ruGPT-3.5, 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=2,
bot_token_id=46787
):
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")
output = model.generate(
prompt=prompt,
generation_config=generation_config
).text
# print("output", output)
output_ids = tokenizer(output, return_tensors="pt")['input_ids'][0]
# print("output_ids", output_ids)
# output_ids = output_ids[len(data["input_ids"][0]):]
# print("output_ids", output_ids)
output = tokenizer.decode(output_ids, skip_special_tokens=True)
# print("output_ids", output)
return output.strip()
# Load base model
model = AutoModel.from_pretrained(
MODEL_NAME,
model_file="ruGPT-3.5-13B-lora-q4_0.bin",
)
# Init basic tokenizer
tokenizer = AutoTokenizer.from_pretrained('ai-forever/ruGPT-3.5-13B', use_fast=False)
generation_config = GenerationConfig.from_pretrained('ai-forever/ruGPT-3.5-13B')
# Instantiate generator config
g_config = GConfig()
g_config.top_p = generation_config['top_p']
g_config.top_k = generation_config['top_k']
g_config.repetition_penalty = generation_config['repetition_penalty']
g_config.temperature = generation_config['temperature']
g_config.max_new_tokens = 200
# Start conversation
conversation = Conversation()
while True:
user_message = input("User: ")
if user_message.strip() == "/reset":
conversation = Conversation()
print("History reset completed!")
continue
if user_message.strip() == "":
continue
conversation.add_user_message(user_message)
prompt = conversation.get_prompt(tokenizer)
output = generate(
model=model,
tokenizer=tokenizer,
prompt=prompt,
generation_config=g_config
)
conversation.add_bot_message(output)
print("ruGPT-3.5:", output)
print()
print("==============================")
print()
Examples
User: Привет! Почему трава зелёная?
ruGPT-3.5: Трава зелёная благодаря хлорофиллу - пигменту растений, который поглощает световые волны определенной длины и преобразует их в энергию химических связей. Хлорофилл также обеспечивает растения питательными веществами из почвы через фотосинтез.
User: Придумай стишок про нейросети.
ruGPT-3.5: Нейросеть – это чудо науки,
Она способна решать любые задачи.
И если ты хочешь стать программистом,
То лучше всего изучать нейронные сети.
User: Высота самой высокой горы на Земле?
ruGPT-3.5: Самая высокая гора на Земле находится в Гималаях и называется Эверест (8848 метров).
User: Сколько максимум лет может прожить галопогосская черепаха?
ruGPT-3.5: Галопогосские черепахи могут жить до 100 лет и более. Они являются одними из самых долгоживущих животных на планете.
📚 Training Datasets
The datasets utilized for training this model are consistent with those used for Saiga-2.
Here's the comprehensive list:
- ru_turbo_alpaca
- ru_turbo_alpaca_evol_instruct
- ru_turbo_saiga
- ru_sharegpt_cleaned
- oasst1_ru_main_branch
- gpt_roleplay_realm
- ru_instruct_gpt4
🛠 Training Procedure
The following bitsandbytes
quantization config was used during training:
- quant_method: bitsandbytes
- load_in_8bit: True
- load_in_4bit: False
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: fp4
- bnb_4bit_use_double_quant: False
- bnb_4bit_compute_dtype: float32
⚙️ Framework Versions
Ensure you have the following framework versions for compatibility:
- PyTorch 2.1.0
- PEFT 0.5.0
- bitsandbytes 0.41.1
- transformers 4.34.0
Links
- Downloads last month
- 37
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.