inference: false
license: other
license_name: mnpl
license_link: https://mistral.ai/licences/MNPL-0.1.md
tags:
- code
language:
- code
Basic Context Obedient Prompt that works great for RAG
Example usecase where a stock model would make things up. https://imgur.com/9Z4o0tq
Ram Usage for Q8 Quantized model on an RTX3090 24GB https://imgur.com/GPlGLme
Contextual-Request:
BEGININPUT
BEGINCONTEXT
date: 2024-05-03
url: https://web.site.thisshitsbadouthereboys/123
ENDCONTEXT
Pandemic Warning Notice there has been a huge issue with Zombie humans that are passing on a new disease that appeared to be similar to the symptoms of covid but when a host dies they reanimate as a zombie corpse.
ENDINPUT
BEGININSTRUCTION
What is the the pandemic about? cite your sources
ENDINSTRUCTION
### Contextual Response:
Overview This model is meant to enhance adherence to provided context (e.g., for RAG applications) and reduce hallucinations, inspired by airoboros context-obedient question answer format.
Overview
The format for a contextual prompt is as follows:
Contextual-Request:
BEGININPUT
BEGINCONTEXT
[key0: value0]
[key1: value1]
... other metdata ... like character Mood: Scared, Tone of the scene: Spooky, anything that will enhance your RAG / experience, maybe even use small Mermaid Knowledge Graphs as core memories stored as events like I do in my assistant I am building.
ENDCONTEXT
[insert your text blocks here, this is where RAG content goes]
ENDINPUT
[add as many other blocks, in the exact same format]
BEGININSTRUCTION
[insert your instruction(s). The model was tuned with single questions, paragraph format, lists, etc.]
ENDINSTRUCTION
I know it's a bit verbose and annoying, but after much trial and error, using these explicit delimiters helps the model understand where to find the responses and how to associate specific sources with it.
Contextual-Request:
- denotes the type of request pattern the model is to follow for consistencyBEGININPUT
- denotes a new input blockBEGINCONTEXT
- denotes the block of context (metadata key/value pairs) to associate with the current input blockENDCONTEXT
- denotes the end of the metadata block for the current input- [text] - Insert whatever text you want for the input block, as many paragraphs as can fit in the context.
ENDINPUT
- denotes the end of the current input block- [repeat as many input blocks in this format as you want]
BEGININSTRUCTION
- denotes the start of the list (or one) instruction(s) to respond to for all of the input blocks above.- [instruction(s)]
ENDINSTRUCTION
- denotes the end of instruction set
Here's a trivial, but important example to prove the point:
Contextual-Request:
BEGININPUT
BEGINCONTEXT
date: 2021-01-01
url: https://web.site/123
ENDCONTEXT
In a shocking turn of events, blueberries are now green, but will be sticking with the same name.
ENDINPUT
BEGININSTRUCTION
What color are bluberries? Source?
ENDINSTRUCTION
And the expected response:
### Contextual Response:
Blueberries are now green.
Source:
date: 2021-01-01
url: https://web.site/123
References in response
As shown in the example, the dataset includes many examples of including source details in the response, when the question asks for source/citation/references.
Why do this? Well, the R in RAG seems to be the weakest link in the chain. Retrieval accuracy, depending on many factors including the overall dataset size, can be quite low. This accuracy increases when retrieving more documents, but then you have the issue of actually using the retrieved documents in prompts. If you use one prompt per document (or document chunk), you know exactly which document the answer came from, so there's no issue. If, however, you include multiple chunks in a single prompt, it's useful to include the specific reference chunk(s) used to generate the response, rather than naively including references to all of the chunks included in the prompt.
For example, suppose I have two documents:
url: http://foo.bar/1
Strawberries are tasty.
url: http://bar.foo/2
The cat is blue.
If the question being asked is What color is the cat?
, I would only expect the 2nd document to be referenced in the response, as the other link is irrelevant.
BASE MODEL: Model Card for Codestral-22B-v0.1
Codestrall-22B-v0.1 is trained on a diverse dataset of 80+ programming languages, including the most popular ones, such as Python, Java, C, C++, JavaScript, and Bash (more details in the Blogpost). The model can be queried:
- As instruct, for instance to answer any questions about a code snippet (write documentation, explain, factorize) or to generate code following specific indications
- As Fill in the Middle (FIM), to predict the middle tokens between a prefix and a suffix (very useful for software development add-ons like in VS Code)
Installation
It is recommended to use mistralai/Codestral-22B-v0.1
with mistral-inference.
pip install mistral_inference
Download
from huggingface_hub import snapshot_download
from pathlib import Path
mistral_models_path = Path.home().joinpath('mistral_models', 'Codestral-22B-v0.1')
mistral_models_path.mkdir(parents=True, exist_ok=True)
snapshot_download(repo_id="mistralai/Codestral-22B-v0.1", allow_patterns=["params.json", "consolidated.safetensors", "tokenizer.model.v3"], local_dir=mistral_models_path)
Chat
After installing mistral_inference
, a mistral-chat
CLI command should be available in your environment.
mistral-chat $HOME/mistral_models/Codestral-22B-v0.1 --instruct --max_tokens 256
Will generate an answer to "Write me a function that computes fibonacci in Rust" and should give something along the following lines:
Sure, here's a simple implementation of a function that computes the Fibonacci sequence in Rust. This function takes an integer `n` as an argument and returns the `n`th Fibonacci number.
fn fibonacci(n: u32) -> u32 {
match n {
0 => 0,
1 => 1,
_ => fibonacci(n - 1) + fibonacci(n - 2),
}
}
fn main() {
let n = 10;
println!("The {}th Fibonacci number is: {}", n, fibonacci(n));
}
This function uses recursion to calculate the Fibonacci number. However, it's not the most efficient solution because it performs a lot of redundant calculations. A more efficient solution would use a loop to iteratively calculate the Fibonacci numbers.
Fill-in-the-middle (FIM)
After installing mistral_inference
and running pip install --upgrade mistral_common
to make sure to have mistral_common>=1.2 installed:
from mistral_inference.model import Transformer
from mistral_inference.generate import generate
from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
from mistral_common.tokens.instruct.request import FIMRequest
tokenizer = MistralTokenizer.v3()
model = Transformer.from_folder("~/codestral-22B-240529")
prefix = """def add("""
suffix = """ return sum"""
request = FIMRequest(prompt=prefix, suffix=suffix)
tokens = tokenizer.encode_fim(request).tokens
out_tokens, _ = generate([tokens], model, max_tokens=256, temperature=0.0, eos_id=tokenizer.instruct_tokenizer.tokenizer.eos_id)
result = tokenizer.decode(out_tokens[0])
middle = result.split(suffix)[0].strip()
print(middle)
Should give something along the following lines:
num1, num2):
# Add two numbers
sum = num1 + num2
# return the sum
Limitations
The Codestral-22B-v0.1 does not have any moderation mechanisms. We're looking forward to engaging with the community on ways to make the model finely respect guardrails, allowing for deployment in environments requiring moderated outputs.
License
Codestral-22B-v0.1 is released under the MNLP-0.1
license.
The Mistral AI Team
Albert Jiang, Alexandre Sablayrolles, Alexis Tacnet, Antoine Roux, Arthur Mensch, Audrey Herblin-Stoop, Baptiste Bout, Baudouin de Monicault, Blanche Savary, Bam4d, Caroline Feldman, Devendra Singh Chaplot, Diego de las Casas, Eleonore Arcelin, Emma Bou Hanna, Etienne Metzger, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Harizo Rajaona, Henri Roussez, Jean-Malo Delignon, Jia Li, Justus Murke, Kartik Khandelwal, Lawrence Stewart, Louis Martin, Louis Ternon, Lucile Saulnier, Lélio Renard Lavaud, Margaret Jennings, Marie Pellat, Marie Torelli, Marie-Anne Lachaux, Marjorie Janiewicz, Mickael Seznec, Nicolas Schuhl, Patrick von Platen, Romain Sauvestre, Pierre Stock, Sandeep Subramanian, Saurabh Garg, Sophia Yang, Szymon Antoniak, Teven Le Scao, Thibaut Lavril, Thibault Schueller, Timothée Lacroix, Théophile Gervet, Thomas Wang, Valera Nemychnikova, Wendy Shang, William El Sayed, William Marshall