You can deploy this model and start earning money today!

wwewtech/
russian-it-community-lora

Name: russian-it-community-lora
Availability: InStock

safetensors

Views

No views yet

🦁 Russian IT Community LoRA Model Zoo

58 pre-trained adapters (55 domain adapters + 3 flagship 7B–8B QLoRA), fine-tuned on the RICC corpus (2.91M messages, 171.5k curated SFT dialogues) for Russian-language IT discourse: backend, DevOps, AI/ML, infrastructure.

Catalog regenerated from the Hub file tree on 2026-09-02. Source of truth: the siblings listing of this repository.

⚡ Quick Start: 3-Line Inference

python

import torch
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Qwen/Qwen2.5-1.5B-Instruct"          # any base model from the catalog
adapter_id = "wwewtech/russian-it-community-lora"
subfolder = "qwen2.5_1.5b_instruct"              # choose from the catalog below

tokenizer = AutoTokenizer.from_pretrained(model_id)
base_model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
model = PeftModel.from_pretrained(base_model, adapter_id, subfolder=subfolder)

inputs = tokenizer("<|user|>\nКак настроить репликацию PostgreSQL?\n<|assistant|>\n", return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

📚 Full Catalog (58 Adapters)

#	Adapter Subfolder	Base Model	Hub Link
01	`bloom_1b7`	`bigscience/bloom-1b7`	`bloom_1b7/`
02	`bloom_560m`	`bigscience/bloom-560m`	`bloom_560m/`
03	`codegen_350m_multi`	`Salesforce/codegen-350M-multi`	`codegen_350m_multi/`
04	`deepseek_coder_1.3b_instruct`	`deepseek-ai/deepseek-coder-1.3b-instruct`	`deepseek_coder_1.3b_instruct/`
05	`deepseek_r1_distill_qwen_1.5b`	`deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B`	`deepseek_r1_distill_qwen_1.5b/`
06	`falcon3_1b_instruct`	`tiiuae/Falcon3-1B-Instruct`	`falcon3_1b_instruct/`
07	`falcon3_3b_instruct`	`tiiuae/Falcon3-3B-Instruct`	`falcon3_3b_instruct/`
08	`gemma_2_2b_it`	`unsloth/gemma-2-2b-it`	`gemma_2_2b_it/`
09	`gpt2_large`	`openai-community/gpt2-large`	`gpt2_large/`
10	`gpt2_medium`	`openai-community/gpt2-medium`	`gpt2_medium/`
11	`granite_3b_code_instruct`	`ibm-granite/granite-3b-code-instruct`	`granite_3b_code_instruct/`
12	`heavyweight_deepseek_r1_7b`	`unsloth/DeepSeek-R1-Distill-Qwen-7B-bnb-4bit`	`heavyweight_deepseek_r1_7b/`
13	`heavyweight_llama3.1_8b`	`unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit`	`heavyweight_llama3.1_8b/`
14	`heavyweight_qwen2.5_coder_7b`	`unsloth/Qwen2.5-Coder-7B-Instruct-bnb-4bit`	`heavyweight_qwen2.5_coder_7b/`
15	`llama_3.2_1b_instruct`	`unsloth/Llama-3.2-1B-Instruct`	`llama_3.2_1b_instruct/`
16	`llama_3.2_3b_instruct`	`unsloth/Llama-3.2-3B-Instruct`	`llama_3.2_3b_instruct/`
17	`minicpm_2b_dpo`	`openbmb/MiniCPM-2B-dpo-bf16`	`minicpm_2b_dpo/`
18	`opt_1.3b`	`facebook/opt-1.3b`	`opt_1.3b/`
19	`opt_125m`	`facebook/opt-125m`	`opt_125m/`
20	`opt_2.7b`	`facebook/opt-2.7b`	`opt_2.7b/`
21	`opt_350m`	`facebook/opt-350m`	`opt_350m/`
22	`phi_1_5`	`microsoft/phi-1_5`	`phi_1_5/`
23	`phi_2`	`microsoft/phi-2`	`phi_2/`
24	`phi_3.5_mini_instruct`	`microsoft/Phi-3.5-mini-instruct`	`phi_3.5_mini_instruct/`
25	`phi_3_mini_4k_instruct`	`microsoft/Phi-3-mini-4k-instruct`	`phi_3_mini_4k_instruct/`
26	`pythia_1.4b`	`EleutherAI/pythia-1.4b-deduped`	`pythia_1.4b/`
27	`pythia_2.8b`	`EleutherAI/pythia-2.8b-deduped`	`pythia_2.8b/`
28	`pythia_410m`	`EleutherAI/pythia-410m-deduped`	`pythia_410m/`
29	`pythia_70m`	`EleutherAI/pythia-70m-deduped`	`pythia_70m/`
30	`qwen1.5_0.5b_chat`	`Qwen/Qwen1.5-0.5B-Chat`	`qwen1.5_0.5b_chat/`
31	`qwen1.5_1.8b_chat`	`Qwen/Qwen1.5-1.8B-Chat`	`qwen1.5_1.8b_chat/`
32	`qwen2.5_0.5b_instruct`	`Qwen/Qwen2.5-0.5B-Instruct`	`qwen2.5_0.5b_instruct/`
33	`qwen2.5_1.5b_instruct`	`Qwen/Qwen2.5-1.5B-Instruct`	`qwen2.5_1.5b_instruct/`
34	`qwen2.5_3b_instruct`	`Qwen/Qwen2.5-3B-Instruct`	`qwen2.5_3b_instruct/`
35	`qwen2.5_coder_0.5b_instruct`	`Qwen/Qwen2.5-Coder-0.5B-Instruct`	`qwen2.5_coder_0.5b_instruct/`
36	`qwen2.5_coder_1.5b_instruct`	`Qwen/Qwen2.5-Coder-1.5B-Instruct`	`qwen2.5_coder_1.5b_instruct/`
37	`qwen2.5_coder_3b_instruct`	`Qwen/Qwen2.5-Coder-3B-Instruct`	`qwen2.5_coder_3b_instruct/`
38	`qwen2.5_math_1.5b_instruct`	`Qwen/Qwen2.5-Math-1.5B-Instruct`	`qwen2.5_math_1.5b_instruct/`
39	`qwen2_0.5b_instruct`	`Qwen/Qwen2-0.5B-Instruct`	`qwen2_0.5b_instruct/`
40	`qwen2_1.5b_instruct`	`Qwen/Qwen2-1.5B-Instruct`	`qwen2_1.5b_instruct/`
41	`rugpt3_large`	`ai-forever/rugpt3large_based_on_gpt2`	`rugpt3_large/`
42	`rugpt3_medium`	`ai-forever/rugpt3medium_based_on_gpt2`	`rugpt3_medium/`
43	`rugpt3_small`	`ai-forever/rugpt3small_based_on_gpt2`	`rugpt3_small/`
44	`russian_it_lora`	`Qwen/Qwen2.5-0.5B-Instruct`	`russian_it_lora/`
45	`sber_mgpt`	`ai-forever/mGPT`	`sber_mgpt/`
46	`smollm2_1.7b_instruct`	`HuggingFaceTB/SmolLM2-1.7B-Instruct`	`smollm2_1.7b_instruct/`
47	`smollm2_135m_instruct`	`HuggingFaceTB/SmolLM2-135M-Instruct`	`smollm2_135m_instruct/`
48	`smollm2_360m_instruct`	`HuggingFaceTB/SmolLM2-360M-Instruct`	`smollm2_360m_instruct/`
49	`smollm_1.7b_instruct`	`HuggingFaceTB/SmolLM-1.7B-Instruct`	`smollm_1.7b_instruct/`
50	`smollm_135m_instruct`	`HuggingFaceTB/SmolLM-135M-Instruct`	`smollm_135m_instruct/`
51	`smollm_360m_instruct`	`HuggingFaceTB/SmolLM-360M-Instruct`	`smollm_360m_instruct/`
52	`stablelm_2_1_6b_chat`	`stabilityai/stablelm-2-1_6b-chat`	`stablelm_2_1_6b_chat/`
53	`stablelm_2_zephyr_1_6b`	`stabilityai/stablelm-2-zephyr-1_6b`	`stablelm_2_zephyr_1_6b/`
54	`tiny_starcoder_py`	`bigcode/tiny_starcoder_py`	`tiny_starcoder_py/`
55	`tinyllama_1.1b_chat`	`TinyLlama/TinyLlama-1.1B-Chat-v1.0`	`tinyllama_1.1b_chat/`
56	`vikhr_llama_3.2_1b`	`Vikhrmodels/Vikhr-Llama-3.2-1B-instruct`	`vikhr_llama_3.2_1b/`
57	`vikhr_qwen_2.5_0.5b`	`Vikhrmodels/Vikhr-Qwen-2.5-0.5B-Instruct`	`vikhr_qwen_2.5_0.5b/`
58	`vikhr_qwen_2.5_1.5b`	`Vikhrmodels/Vikhr-Qwen-2.5-1.5B-Instruct`	`vikhr_qwen_2.5_1.5b/`

🥇 Flagship QLoRA Models (7B–8B)

Full-precision copies also live under models/: models/heavyweight_qwen2.5_coder_7b, models/heavyweight_deepseek_r1_7b, models/heavyweight_llama3.1_8b.

📓 Training Data & Evaluation Status

Training corpus: RICC SFT Dialogues (171,520 multi-turn dialogues).
Academic benchmark numbers (HumanEval / RuMMLU / PPL) published earlier are withdrawn pending re-evaluation: the harness had answer-parsing and column-mapping defects that produced implausible values (see repo commit history). Enterprise scenario scores are rubric-based heuristics, not capability measurements.