id: devstral-small-2-24b-instruct-2512-q4_k_m
name: Devstral Small 2 24B Instruct 2512 GGUF Q4_K_M
type: gguf-model
task:
  - 1c
  - code
  - agentic-coding
  - text
  - tool-use
language:
  - ru
  - en
source: huggingface
upstream_id: bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF
upstream_url: https://huggingface.co/bartowski/mistralai_Devstral-Small-2-24B-Instruct-2512-GGUF
license: apache-2.0
status: candidate
created_at: 2026-06-18
storage_path: /models/gguf/1c/devstral-small-2-24b-instruct-2512-q4_k_m
filename: mistralai_Devstral-Small-2-24B-Instruct-2512-Q4_K_M.gguf
file_size_bytes: 14334438272
format: gguf
quantization: Q4_K_M
context_length: null
deployment_context_length: 32768
vram_required_gb: null
parameters: 24B
base_model: mistralai/Devstral-Small-2-24B-Instruct-2512
adapter_for: null
datasets: []
eval_suites:
  - plugins/1c/evals/smoke.yaml
deployment:
  target: docker-gpu.cin.su
  runtime: llama.cpp
  served_model_name: devstral-1c-q4
  compose: core/deploy/docker-gpu/llama-cpp/compose.yaml
notes: "GGUF Q4_K_M quant for 1C/code experiments. Loads on docker-gpu.cin.su with llama.cpp CUDA and uses GPU, but current llama.cpp server returns empty decoded content for chat/completion despite token generation. Runtime is experimental until a compatible template/runtime/quant is found."