README.md

metadata

license: apache-2.0
base_model:
  - mistralai/Devstral-Small-2507
language:
  - en
  - fr
  - de
  - es
  - pt
  - it
  - ja
  - ko
  - ru
  - zh
  - ar
  - fa
  - id
  - ms
  - ne
  - pl
  - ro
  - sr
  - sv
  - tr
  - uk
  - vi
  - hi
  - bn
pipeline_tag: text-generation
tags:
  - merge
  - programming
  - code generation
  - code
  - coding
  - coder
  - chat
  - code
  - chat
  - brainstorm
  - brainstorm20x
  - mistral
library_name: transformers

Mistral-Devstral-2507-CODER-Brainstorm20x-34B

This repo contains the full precision source code, in "safe tensors" format to generate GGUFs, GPTQ, EXL2, AWQ, HQQ and other formats. The source code can also be used directly.

This model contains Brainstorm 20x, combined with Mistral's 24B Coder (instruct model):

https://huggingface.co/mistralai/Devstral-Small-2507

Information on the 24B Mistral model below, followed by Brainstorm 20x adapter (by DavidAU) and then a complete help section for running LLM / AI models.

The Brainstorm adapter improves code generation, and unique code solving abilities.

This model requires:

• Jinja (embedded) or CHATML template
• Max context of 128k.

Settings used for testing (suggested):

• Temp .3 to .7
• Rep pen 1.05 to 1.1
• Topp .8 , minp .05
• Topk 20
• No system prompt.

This model will respond well to both detailed instructions and step by step refinement and additions to code.

As this is an instruct model, it will also benefit from a detailed system prompt too.

For simpler coding problems, lower quants will work well; but for complex/multi-step problem solving suggest Q6 or Q8.

---

Devstral Small 1.1

Devstral is an agentic LLM for software engineering tasks built under a collaboration between Mistral AI and All Hands AI 🙌. Devstral excels at using tools to explore codebases, editing multiple files and power software engineering agents. The model achieves remarkable performance on SWE-bench which positions it as the #1 open source model on this benchmark.

It is finetuned from Mistral-Small-3.1, therefore it has a long context window of up to 128k tokens. As a coding agent, Devstral is text-only and before fine-tuning from Mistral-Small-3.1 the vision encoder was removed.

For enterprises requiring specialized capabilities (increased context, domain-specific knowledge, etc.), we will release commercial models beyond what Mistral AI contributes to the community.

Learn more about Devstral in our blog post.

Updates compared to Devstral Small 1.0:

• Improved performance, please refer to the benchmark results.
• Devstral Small 1.1 is still great when paired with OpenHands. This new version also generalizes better to other prompts and coding environments.
• Supports Mistral's function calling format.

Key Features:

• Agentic coding: Devstral is designed to excel at agentic coding tasks, making it a great choice for software engineering agents.
• lightweight: with its compact size of just 24 billion parameters, Devstral is light enough to run on a single RTX 4090 or a Mac with 32GB RAM, making it an appropriate model for local deployment and on-device use.
• Apache 2.0 License: Open license allowing usage and modification for both commercial and non-commercial purposes.
• Context Window: A 128k context window.
• Tokenizer: Utilizes a Tekken tokenizer with a 131k vocabulary size.

Benchmark Results

SWE-Bench

Devstral Small 1.1 achieves a score of 53.6% on SWE-Bench Verified, outperforming Devstral Small 1.0 by +6,8% and the second best state of the art model by +11.4%.

Model	Agentic Scaffold	SWE-Bench Verified (%)
Devstral Small 1.1	OpenHands Scaffold	53.6
Devstral Small 1.0	OpenHands Scaffold	46.8
GPT-4.1-mini	OpenAI Scaffold	23.6
Claude 3.5 Haiku	Anthropic Scaffold	40.6
SWE-smith-LM 32B	SWE-agent Scaffold	40.2
Skywork SWE	OpenHands Scaffold	38.0
DeepSWE	R2E-Gym Scaffold	42.2

When evaluated under the same test scaffold (OpenHands, provided by All Hands AI 🙌), Devstral exceeds far larger models such as Deepseek-V3-0324 and Qwen3 232B-A22B.

Usage

We recommend to use Devstral with the OpenHands scaffold. You can use it either through our API or by running locally.

API

Follow these instructions to create a Mistral account and get an API key.

Then run these commands to start the OpenHands docker container.

export MISTRAL_API_KEY=<MY_KEY>

mkdir -p ~/.openhands && echo '{"language":"en","agent":"CodeActAgent","max_iterations":null,"security_analyzer":null,"confirmation_mode":false,"llm_model":"mistral/devstral-small-2507","llm_api_key":"'$MISTRAL_API_KEY'","remote_runtime_resource_factor":null,"github_token":null,"enable_default_condenser":true}' > ~/.openhands-state/settings.json

docker pull docker.all-hands.dev/all-hands-ai/runtime:0.48-nikolaik

docker run -it --rm --pull=always \
    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.48-nikolaik \
    -e LOG_ALL_EVENTS=true \
    -v /var/run/docker.sock:/var/run/docker.sock \
    -v ~/.openhands:/.openhands \
    -p 3000:3000 \
    --add-host host.docker.internal:host-gateway \
    --name openhands-app \
    docker.all-hands.dev/all-hands-ai/openhands:0.48

Local inference

The model can also be deployed with the following libraries:

• vllm (recommended): See here
• mistral-inference: See here
• transformers: See here
• LMStudio: See here
• llama.cpp: See here
• ollama: See here

vLLM (recommended)

Expand

We recommend using this model with the vLLM library to implement production-ready inference pipelines.

Installation

Make sure you install vLLM >= 0.9.1:

pip install vllm --upgrade

Also make sure to have installed mistral_common >= 1.7.0.

pip install mistral-common --upgrade

To check:

python -c "import mistral_common; print(mistral_common.__version__)"

You can also make use of a ready-to-go docker image or on the docker hub.

Launch server

We recommand that you use Devstral in a server/client setting.

1. Spin up a server:

vllm serve mistralai/Devstral-Small-2507 --tokenizer_mode mistral --config_format mistral --load_format mistral --tool-call-parser mistral --enable-auto-tool-choice --tensor-parallel-size 2

2. To ping the client you can use a simple Python snippet.

import requests
import json
from huggingface_hub import hf_hub_download


url = "http://<your-server-url>:8000/v1/chat/completions"
headers = {"Content-Type": "application/json", "Authorization": "Bearer token"}

model = "mistralai/Devstral-Small-2507"

def load_system_prompt(repo_id: str, filename: str) -> str:
    file_path = hf_hub_download(repo_id=repo_id, filename=filename)
    with open(file_path, "r") as file:
        system_prompt = file.read()
    return system_prompt

SYSTEM_PROMPT = load_system_prompt(model, "SYSTEM_PROMPT.txt")

messages = [
    {"role": "system", "content": SYSTEM_PROMPT},
    {
        "role": "user",
        "content": [
            {
                "type": "text",
                "text": "<your-command>",
            },
        ],
    },
]

data = {"model": model, "messages": messages, "temperature": 0.15}

# Devstral Small 1.1 supports tool calling. If you want to use tools, follow this:
# tools = [ # Define tools for vLLM
#     {
#         "type": "function",
#         "function": {
#             "name": "git_clone",
#             "description": "Clone a git repository",
#             "parameters": {
#                 "type": "object",
#                 "properties": {
#                     "url": {
#                         "type": "string",
#                         "description": "The url of the git repository",
#                     },
#                 },
#                 "required": ["url"],
#             },
#         },
#     }
# ] 
# data = {"model": model, "messages": messages, "temperature": 0.15, "tools": tools} # Pass tools to payload.

response = requests.post(url, headers=headers, data=json.dumps(data))
print(response.json()["choices"][0]["message"]["content"])

Mistral-inference

Expand

We recommend using mistral-inference to quickly try out / "vibe-check" Devstral.

Installation

Make sure to have mistral_inference >= 1.6.0 installed.

pip install mistral_inference --upgrade

Download

from huggingface_hub import snapshot_download
from pathlib import Path

mistral_models_path = Path.home().joinpath('mistral_models', 'Devstral')
mistral_models_path.mkdir(parents=True, exist_ok=True)

snapshot_download(repo_id="mistralai/Devstral-Small-2507", allow_patterns=["params.json", "consolidated.safetensors", "tekken.json"], local_dir=mistral_models_path)

Chat

You can run the model using the following command:

mistral-chat $HOME/mistral_models/Devstral --instruct --max_tokens 300

You can then prompt it with anything you'd like.

Transformers

Expand

To make the best use of our model with transformers make sure to have installed mistral-common >= 1.7.0 to use our tokenizer.

pip install mistral-common --upgrade

Then load our tokenizer along with the model and generate:

import torch

from mistral_common.protocol.instruct.messages import (
    SystemMessage, UserMessage
)
from mistral_common.protocol.instruct.request import ChatCompletionRequest
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
from huggingface_hub import hf_hub_download
from transformers import AutoModelForCausalLM

def load_system_prompt(repo_id: str, filename: str) -> str:
    file_path = hf_hub_download(repo_id=repo_id, filename=filename)
    with open(file_path, "r") as file:
        system_prompt = file.read()
    return system_prompt

model_id = "mistralai/Devstral-Small-2507"
SYSTEM_PROMPT = load_system_prompt(model_id, "SYSTEM_PROMPT.txt")


tokenizer = MistralTokenizer.from_hf_hub(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

tokenized = tokenizer.encode_chat_completion(
    ChatCompletionRequest(
        messages=[
            SystemMessage(content=SYSTEM_PROMPT),
            UserMessage(content="<your-command>"),
        ],
    )
)

output = model.generate(
    input_ids=torch.tensor([tokenized.tokens]),
    max_new_tokens=1000,
)[0]

decoded_output = tokenizer.decode(output[len(tokenized.tokens):])
print(decoded_output)

LM Studio

Expand

Download the weights from either:

• LM Studio GGUF repository (recommended): https://huggingface.co/lmstudio-community/Devstral-Small-2507-GGUF
• our GGUF repository: https://huggingface.co/mistralai/Devstral-Small-2507_gguf

pip install -U "huggingface_hub[cli]"
huggingface-cli download \
"lmstudio-community/Devstral-Small-2507-GGUF" \ # or mistralai/Devstral-Small-2507_gguf
--include "Devstral-Small-2507-Q4_K_M.gguf" \
--local-dir "Devstral-Small-2507_gguf/"

You can serve the model locally with LMStudio.

• Download LM Studio and install it
• Install lms cli ~/.lmstudio/bin/lms bootstrap
• In a bash terminal, run lms import Devstral-Small-2507-Q4_K_M.gguf in the directory where you've downloaded the model checkpoint (e.g. Devstral-Small-2507_gguf)
• Open the LM Studio application, click the terminal icon to get into the developer tab. Click select a model to load and select Devstral Small 2507. Toggle the status button to start the model, in setting toggle Serve on Local Network to be on.
• On the right tab, you will see an API identifier which should be devstral-small-2507 and an api address under API Usage. Keep note of this address, this is used for OpenHands or Cline.

llama.cpp

Expand

Download the weights from huggingface:

pip install -U "huggingface_hub[cli]"
huggingface-cli download \
"mistralai/Devstral-Small-2507_gguf" \
--include "Devstral-Small-2507-Q4_K_M.gguf" \
--local-dir "mistralai/Devstral-Small-2507_gguf/"

Then run Devstral using the llama.cpp server.

./llama-server -m mistralai/Devstral-Small-2507_gguf/Devstral-Small-2507-Q4_K_M.gguf -c 0 # -c configure the context size, 0 means model's default, here 128k.

OpenHands (recommended)

Launch a server to deploy Devstral Small 1.1

Make sure you launched an OpenAI-compatible server such as vLLM or Ollama as described above. Then, you can use OpenHands to interact with Devstral Small 1.1.

In the case of the tutorial we spineed up a vLLM server running the command:

vllm serve mistralai/Devstral-Small-2507 --tokenizer_mode mistral --config_format mistral --load_format mistral --tool-call-parser mistral --enable-auto-tool-choice --tensor-parallel-size 2

The server address should be in the following format: http://<your-server-url>:8000/v1

Launch OpenHands

You can follow installation of OpenHands here.

The easiest way to launch OpenHands is to use the Docker image:

docker pull docker.all-hands.dev/all-hands-ai/runtime:0.48-nikolaik

docker run -it --rm --pull=always \
    -e SANDBOX_RUNTIME_CONTAINER_IMAGE=docker.all-hands.dev/all-hands-ai/runtime:0.48-nikolaik \
    -e LOG_ALL_EVENTS=true \
    -v /var/run/docker.sock:/var/run/docker.sock \
    -v ~/.openhands:/.openhands \
    -p 3000:3000 \
    --add-host host.docker.internal:host-gateway \
    --name openhands-app \
    docker.all-hands.dev/all-hands-ai/openhands:0.48

Then, you can access the OpenHands UI at http://localhost:3000.

Connect to the server

When accessing the OpenHands UI, you will be prompted to connect to a server. You can use the advanced mode to connect to the server you launched earlier.

Fill the following fields:

• Custom Model: openai/mistralai/Devstral-Small-2507
• Base URL: http://<your-server-url>:8000/v1
• API Key: token (or any other token you used to launch the server if any)

See settings

Cline

Launch a server to deploy Devstral Small 1.1

Make sure you launched an OpenAI-compatible server such as vLLM or Ollama as described above. Then, you can use OpenHands to interact with Devstral Small 1.1.

In the case of the tutorial we spineed up a vLLM server running the command:

vllm serve mistralai/Devstral-Small-2507 --tokenizer_mode mistral --config_format mistral --load_format mistral --tool-call-parser mistral --enable-auto-tool-choice --tensor-parallel-size 2

The server address should be in the following format: http://<your-server-url>:8000/v1

Launch Cline

You can follow installation of Cline here. Then you can configure the server address in the settings.

See settings

See more here:

https://huggingface.co/mistralai/Devstral-Small-2507

---

What is Brainstorm?

---

Brainstorm 20x

The BRAINSTORM process was developed by David_AU.

Some of the core principals behind this process are discussed in this scientific paper : Progressive LLaMA with Block Expansion .

However I went in a completely different direction from what was outlined in this paper.

What is "Brainstorm" ?

The reasoning center of an LLM is taken apart, reassembled, and expanded.

In this case for this model: 20 times

Then these centers are individually calibrated. These "centers" also interact with each other. This introduces subtle changes into the reasoning process. The calibrations further adjust - dial up or down - these "changes" further. The number of centers (5x,10x etc) allow more "tuning points" to further customize how the model reasons so to speak.

The core aim of this process is to increase the model's detail, concept and connection to the "world", general concept connections, prose quality and prose length without affecting instruction following.

This will also enhance any creative use case(s) of any kind, including "brainstorming", creative art form(s) and like case uses.

Here are some of the enhancements this process brings to the model's performance:

• Prose generation seems more focused on the moment to moment.
• Sometimes there will be "preamble" and/or foreshadowing present.
• Fewer or no "cliches"
• Better overall prose and/or more complex / nuanced prose.
• A greater sense of nuance on all levels.
• Coherence is stronger.
• Description is more detailed, and connected closer to the content.
• Simile and Metaphors are stronger and better connected to the prose, story, and character.
• Sense of "there" / in the moment is enhanced.
• Details are more vivid, and there are more of them.
• Prose generation length can be long to extreme.
• Emotional engagement is stronger.
• The model will take FEWER liberties vs a normal model: It will follow directives more closely but will "guess" less.
• The MORE instructions and/or details you provide the more strongly the model will respond.
• Depending on the model "voice" may be more "human" vs original model's "voice".

Other "lab" observations:

• This process does not, in my opinion, make the model 5x or 10x "smarter" - if only that was true!
• However, a change in "IQ" was not an issue / a priority, and was not tested or calibrated for so to speak.
• From lab testing it seems to ponder, and consider more carefully roughly speaking.
• You could say this process sharpens the model's focus on it's task(s) at a deeper level.

The process to modify the model occurs at the root level - source files level. The model can quanted as a GGUF, EXL2, AWQ etc etc.

---

For more information / other Qwen/Mistral Coders / additional settings see:

[ https://huggingface.co/DavidAU/Qwen2.5-MOE-2x-4x-6x-8x__7B__Power-CODER__19B-30B-42B-53B-gguf ]

---

Help, Adjustments, Samplers, Parameters and More

---

CHANGE THE NUMBER OF ACTIVE EXPERTS:

See this document:

https://huggingface.co/DavidAU/How-To-Set-and-Manage-MOE-Mix-of-Experts-Model-Activation-of-Experts

Settings: CHAT / ROLEPLAY and/or SMOOTHER operation of this model:

In "KoboldCpp" or "oobabooga/text-generation-webui" or "Silly Tavern" ;

Set the "Smoothing_factor" to 1.5

: in KoboldCpp -> Settings->Samplers->Advanced-> "Smooth_F"

: in text-generation-webui -> parameters -> lower right.

: In Silly Tavern this is called: "Smoothing"

NOTE: For "text-generation-webui"

-> if using GGUFs you need to use "llama_HF" (which involves downloading some config files from the SOURCE version of this model)

Source versions (and config files) of my models are here:

https://huggingface.co/collections/DavidAU/d-au-source-files-for-gguf-exl2-awq-gptq-hqq-etc-etc-66b55cb8ba25f914cbf210be

OTHER OPTIONS:

• Increase rep pen to 1.1 to 1.15 (you don't need to do this if you use "smoothing_factor")

• If the interface/program you are using to run AI MODELS supports "Quadratic Sampling" ("smoothing") just make the adjustment as noted.

Highest Quality Settings / Optimal Operation Guide / Parameters and Samplers

This a "Class 1" model:

For all settings used for this model (including specifics for its "class"), including example generation(s) and for advanced settings guide (which many times addresses any model issue(s)), including methods to improve model performance for all use case(s) as well as chat, roleplay and other use case(s) please see:

[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]

You can see all parameters used for generation, in addition to advanced parameters and samplers to get the most out of this model here:

[ https://huggingface.co/DavidAU/Maximizing-Model-Performance-All-Quants-Types-And-Full-Precision-by-Samplers_Parameters ]