Machinelearning
Погружаемся в машинное обучение и Data Science Показываем как запускать любые LLm на пальцах. По всем вопросам - @haarrp @itchannels_telegram -🔥best channels Реестр РКН: clck.ru/3Fmqri
Show more📈 Analytical overview of Telegram channel Machinelearning
Channel Machinelearning (@ai_machinelearning_big_data) in the Russian language segment is an active participant. Currently, the community unites 295 549 subscribers, ranking 332 in the Technologies & Applications category and 1 273 in the Russia region.
📊 Audience metrics and dynamics
Since its creation on невідомо, the project has demonstrated rapid growth, gathering an audience of 295 549 subscribers.
According to the latest data from 23 June, 2026, the channel demonstrates stable activity. Although there has been a change in the number of participants by -6 330 over the last 30 days and by -217 over the last 24 hours, overall reach remains high.
- Verification status: Not verified
- Engagement rate (ER): The average audience engagement rate is 7.94%. Within the first 24 hours after publication, content typically collects 5.68% reactions from the total number of subscribers.
- Post reach: On average, each post receives 23 490 views. Within the first day, a publication typically gains 16 791 views.
- Reactions and interaction: The audience actively supports content: the average number of reactions per post is 190.
- Thematic interests: Content is focused on key topics such as openai, claude, api, gemini, контекст.
📝 Description and content policy
The author describes the resource as a platform for expressing subjective opinions:
“Погружаемся в машинное обучение и Data Science
Показываем как запускать любые LLm на пальцах.
По всем вопросам - @haarrp
@itchannels_telegram -🔥best channels
Реестр РКН: clck.ru/3Fmqri”
Thanks to the high frequency of updates (latest data received on 24 June, 2026), the channel maintains relevance and a high level of publication reach. Analytics show that the audience actively interacts with content, making it an important point of influence in the Technologies & Applications category.
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained(
"LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct",
torch_dtype=torch.bfloat16,
trust_remote_code=True,
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct")
# Choose your prompt
prompt = "Explain who you are" # English example
prompt = "너의 소원을 말해봐" # Korean example
messages = [
{"role": "system", "content": "You are EXAONE model from LG AI Research, a helpful assistant."},
{"role": "user", "content": prompt}
]
input_ids = tokenizer.apply_chat_template(
messages,
tokenize=True,
add_generation_prompt=True,
return_tensors="pt"
)
output = model.generate(
input_ids.to("cuda"),
eos_token_id=tokenizer.eos_token_id,
max_new_tokens=128
)
print(tokenizer.decode(output[0]))
📌Лицензирование : использование разрешено исключительно в некоммерческих целях. Любое коммерческое использование модели требует отдельной лицензии от правообладателя.
🟡Страница проекта
🟡Arxiv
🟡Модель на HF
🟡Demo
🖥Github [ Stars: 123 | Issues: 0 | Forks: 5]
@ai_machinelearning_big_data
#AI #LLM #ML #EXAONE #LG# Open command prompt and run
git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git
webui-user.bat
# Put downloaded models from HF into models/StableDiffusion
📌Лицензирование : AGPL-3.0 license
🟡Модель Flux-dev-NF4
🟡Модель Flux-dev-FP8
🖥Github [ Stars: 5.8K | Issues: 405 | Forks: 580]
@ai_machinelearning_big_data
#AI #Forge #ML #FLUX# Clone repository and install dependences:
pip install git+https://github.com/huggingface/parler-tts.git
# Inference with random voice
import torch
from parler_tts import ParlerTTSForConditionalGeneration
from transformers import AutoTokenizer
import soundfile as sf
device = "cuda:0" if torch.cuda.is_available() else "cpu"
model = ParlerTTSForConditionalGeneration.from_pretrained("parler-tts/parler-tts-mini-v1").to(device)
tokenizer = AutoTokenizer.from_pretrained("parler-tts/parler-tts-mini-v1")
prompt = "Hey, how are you doing today?"
description = "A female speaker delivers a slightly expressive and animated speech with a moderate speed and pitch. The recording is of very high quality, with the speaker's voice sounding clear and very close up."
input_ids = tokenizer(description, return_tensors="pt").input_ids.to(device)
prompt_input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
generation = model.generate(input_ids=input_ids, prompt_input_ids=prompt_input_ids)
audio_arr = generation.cpu().numpy().squeeze()
sf.write("parler_tts_out.wav", audio_arr, model.config.sampling_rate)y
📌Лицензирование : Apache-2.0 license
🟡Модель Parler-TTS Mini
🟡Модель Parler-TTS Large
🟡Arxiv
🟡Demo Video
🟡Google Collab (файнтюн)
🟡Demo
🖥Github [ Stars: 3.4K | Issues: 49 | Forks: 338]
@ai_machinelearning_big_data
#AI #Parler #ML #TTS# Clone repository:
git clone https://github.com/TheMody/No-learning-rates-needed-Introducing-SALSA-Stable-Armijo-Line-Search-Adaptation.git
# Create & activate env:
conda env create -f environment.yml
conda activate sls3
# Install dependencies:
pip install pytorch numpy transformers datasets tensorflow-datasets wandb
# NOTE: custom optimizer is in \salsa\SaLSA.py,comparison version are in \salsa\adam_sls.py:
from salsa.SaLSA import SaLSA
self.optimizer = SaLSA(model.parameters())
# NOTE: typical pytorch forward pass needs to be changed to:
def closure(backwards = False):
y_pred = model(x)
loss = criterion(y_pred, y)
if backwards: loss.backward()
return loss
optimizer.zero_grad()
loss = optimizer.step(closure = closure)
📌Лицензирование : MIT License
🟡Arxiv
🟡Датасет Cifar-10
🟡Youtube video
🖥Github [ Stars: 11 | Issues: 0 | Forks: 0]
@ai_machinelearning_big_data
#AI #LLM #ML #Train #SALSA# Ensure you have latest Hugging face transformers
pip install git+https://github.com/huggingface/transformers
# to build a web UI demoinstall the following packages
pip install -r requirements_web_demo.txt
# run Gradio web UI
python demo/web_demo_audio.py
📌Лицензирование : Apache 2.0
🟡Страница проекта
🟡Коллекция моделей на HF
🟡Arxiv
🟡Сообщество в Discord
🟡Demo
🖥Github [ Stars: 618 | Issues: 7 | Forks: 17]
@ai_machinelearning_big_data
#AI #LLM #ML #Qwen2accelerate launch train_flux_lora_deepspeed.py --config "train_configs/test_lora.yaml"
ControlNet for FLUX dev
accelerate launch train_flux_deepspeed_controlnet.py --config "train_configs/test_canny_controlnet.yaml"
В ближайших планах публикация весов ControlNet для FLUX:
🟢OpenPose
🟢Depth
🟢IP-Adapters
*️⃣RealismLoRA *️⃣ Canny ControlNet для FLUX *️⃣Воркфлоу с поддержкой LoRA для ComfyUI *️⃣Попробовать LoRA онлайн
▶️SimpleTuner также добавил в пакет скриптов поддержку LoRA for FLUX и скрипт обучения для квантованных моделей FLUX int8, int4, int2, fp8.
Рекомендации по ресурсам для LoRA:
🟠Rank-16 LoRA использует чуть больше 40 ГБ VRAM;
🟠GPU AMD и Apple не подходят для обучения Flux.
Наблюдения, сделанные автором SimpleTuner в ходе экспериментов:
🟠Для обучение под Schnell нужно больше времени для тренировки, результаты пока не очень;
🟠LoRA, обученная на dev отлично работает и на Schnell;
🟠Мердж 50/50 моделей dev и Schnell работает, на этом мердже можно тренировать LoRA`s;
🟠Квантованные версии FLUX позволяют использовать оптимизаторы - Prodigy, Adafactor, Dadaptation, AdamW, и AdamW8Bit;
🟠Квантование fp8 выполняется медленнее, чем int8, и может иметь худший результат из-за использования e4m3fn в Quanto;
🟠Плохое качество датасета, слишком высокий LR, неправильный выбор оптимизатора, низкое значение Network при большом датасете, использование нестандартных размеров изображений в датасете - этот все приводит к чудовищным артефактам "квадратной решетки" в результате.
🖥Github ХlabsAI [ Stars: 266 | Issues: 9 | Forks: 12]
🖥Github SimpleTuner [ Stars: 885K | Issues: 13 | Forks: 61]
@ai_machinelearning_big_data
#AI #FLUX #ML #Train #LoRA# Running Test Cases:
> pytest # will run all test cases - including ones that require a gpu
> pytest -m "not gpu" # run test cases that can work with just cpu
# Download the models:
curl https://docs-assets.developer.apple.com/ml-research/models/mdm/flickr64/vis_model.pth --output vis_model_64x64.pth
curl https://docs-assets.developer.apple.com/ml-research/models/mdm/flickr256/vis_model.pth --output vis_model_256x256.pth
curl https://docs-assets.developer.apple.com/ml-research/models/mdm/flickr1024/vis_model.pth --output vis_model_1024x1024.pth
# Launch Web Demo:
torchrun --standalone --nproc_per_node=1 ml_mdm/clis/generate_sample.py --port 19999
⚠️ В Issues репозитория есть обращение о некорректной команде запуска Web Demo. Следите за обновлением тикета и коммитами.
📌Лицензирование : Apple Inc.
🟡Arxiv
🟡Arxiv
🖥Github [ Stars: 11 | Issues: 0 | Forks: 0]
@ai_machinelearning_big_data
#AI #Diffusion #ML #Text2Image #Applemessages = [
{"role": "system", "content": "You are an assistant who gives helpful, detailed, and polite answers to the user's questions based on the context with appropriate reasoning as required. Indicate when the answer cannot be found in the context."},
{"role": "user", "content": """Context: <CONTEXT INFORMATION> \n\n <USER QUERY>"""},
]
📌Лицензирование : Apache-2.0
🟡Страница проекта
🟡Коллекция моделей на HF
@ai_machinelearning_big_data
#AI #LLM #ML #BRAG #RAG# Clone this repository and navigate to the source folder:
git clone https://github.com/OpenBMB/MiniCPM-V.git
cd MiniCPM-V
# Create conda environment:
conda create -n MiniCPM-V python=3.10 -y
conda activate MiniCPM-V
#Install dependencies.
pip install -r requirements.txt
## For NVIDIA GPUs, run::
python web_demo_2.6.py --device cuda
📌Лицензирование:
🟢код - Apache-2.0;
🟠модели - свободно для любых академических исследований. Коммерция - соблюдение этого соглашения.
🟡Tech Report MiniCPM-Llama3-V 2.5
🟡Коллекция моделей на HF
🟡Demo MiniCPM-V 2.6
🟡Demo MiniCPM-Llama3-V 2.5
🟡Demo MiniCPM-V 2
🖥Github [ Stars: 8.3K | Issues: 27 | Forks: 583]
@ai_machinelearning_big_data
#AI #MLLM #ML #MiniCPM #MobileVLM
Available now! Telegram Research 2025 — the year's key insights 
