---
language: ru
library_name: mlx
pipeline_tag: text-generation
tags:
- mlx
- gguf
- safetensors
- qwen2
- qwen2.5-coder
- bitrix
- 1c-bitrix
- qlora
- text-generation-inference
- endpoints_compatible
- 8-bit
---

# BX-GURU v6 — Qwen2.5-Coder-14B-Bitrix-Developer

**Подробная статья на русском:** [bazarow.ru/blog-note/16997/](https://bazarow.ru/blog-note/16997/) — история создания, 5 версий Devstral, почему перешли на Qwen.

QLoRA fine-tune of Qwen2.5-Coder-14B-Instruct (8-bit). Specialized in 1C-Bitrix BUS: D7 ORM, events, components, Highload-blocks, Router, REST API, agents, SQL.

## Formats

| Format | File | Usage |
|--------|------|-------|
| 🍎 **MLX** (safetensors) | `model-*-of-*.safetensors` | Mac Apple Silicon (mlx-lm, LM Studio MLX) |
| 🏭 **GGUF** Q8_0 | `bxguru_v6_Q8_0.gguf` | LM Studio GGUF, Ollama, llama.cpp |

## How to use

### LM Studio

Search `MihailBazarov/qwen2.5-coder-14b-bitrix-developer` in LM Studio. Both MLX and GGUF versions are available:

- **MLX** → uses `model-*-of-*.safetensors` (native Apple Silicon)
- **GGUF** → uses `bxguru_v6_Q8_0.gguf` (broader compatibility)

System prompt: *"Ты — BX-GURU, Senior Bitrix fullstack разработчик. Знаешь D7 ORM, компоненты, события, модули, AJAX-контроллеры, инфоблоки, Highload-блоки, Router, REST API, админку."*

### Ollama

```bash
ollama create bxguru -f - << 'EOF'
FROM ./bxguru_v6_Q8_0.gguf
SYSTEM "Ты — BX-GURU, Senior Bitrix fullstack разработчик."
EOF
ollama run bxguru
```

## Capabilities

- D7 ORM: `ElementTable::getList()`, `SectionTable`, filters, sorting, `ExpressionField`, `ReferenceField`
- Events: `EventManager::addEventHandler()`, `OnBefore/OnAfter` handlers
- Highload-blocks: `HighloadBlockTable::compileEntity()`
- REST API, routing, AJAX controllers
- Agents & CRON: `CAgent::AddAgent()`
- SQL: JOIN, GROUP BY, subqueries

## Links

- Source code & training guide: [gitverse](https://gitverse.ru/mibazarow/devstral-small-24b-senior-bitrix-developer)