Mistral 7B

Abstract

We introduce Mistral 7B, a 7–billion-parameter language model engineered for superior performance and efficiency. Mistral 7B outperforms the best open 13B model (Llama 2) across all evaluated benchmarks, and the best released 34B model (Llama 1) in reasoning, mathematics, and code generation. Our model leverages grouped-query attention (GQA) for faster inference, coupled with sliding window attention (SWA) to effectively handle sequences of arbitrary length with a reduced inference cost. We also provide a model fine-tuned to follow instructions, Mistral 7B – Instruct, that surpasses Llama 2 13B – chat model both on human and automated benchmarks. Our models are released under the Apache 2.0 license.
Code: https://github.com/mistralai/mistral-src
Webpage: https://mistral.ai/news/announcing-mistral-7b/

1 Introduction

In the rapidly evolving domain of Natural Language Processing (NLP), the race towards higher model performance often necessitates an escalation in model size. However, this scaling tends to increase computational costs and inference latency, thereby raising barriers to deployment in practical, real-world scenarios. In this context, the search for balanced models delivering both high-level performance and efficiency becomes critically essential. Our model, Mistral 7B, demonstrates that a carefully designed language model can deliver high performance while maintaining an efficient inference. Mistral 7B outperforms the previous best 13B model (Llama 2, touvron2023llama2 ) across all tested benchmarks, and surpasses the best 34B model (LLaMa 34B, touvron2023llama ) in mathematics and code generation. Furthermore, Mistral 7B approaches the coding performance of Code-Llama 7B roziere2023code , without sacrificing performance on non-code related benchmarks.

Mistral 7B leverages grouped-query attention (GQA) ainslie2023gqa , and sliding window attention (SWA) child2019generating ; beltagy2020longformer . GQA significantly accelerates the inference speed, and also reduces the memory requirement during decoding, allowing for higher batch sizes hence higher throughput, a crucial factor for real-time applications. In addition, SWA is designed to handle longer sequences more effectively at a reduced computational cost, thereby alleviating a common limitation in LLMs. These attention mechanisms collectively contribute to the enhanced performance and efficiency of Mistral 7B.

Mistral 7B is released under the Apache 2.0 license. This release is accompanied by a reference implementation facilitating easy deployment either locally or on cloud platforms such as AWS, GCP, or Azure using the vLLM kwon2023efficient inference server and SkyPilot . Integration with Hugging Face  is also streamlined for easier integration. Moreover, Mistral 7B is crafted for ease of fine-tuning across a myriad of tasks. As a demonstration of its adaptability and superior performance, we present a chat model fine-tuned from Mistral 7B that significantly outperforms the Llama 2 13B – Chat model.

Mistral 7B takes a significant step in balancing the goals of getting high performance while keeping large language models efficient. Through our work, our aim is to help the community create more affordable, efficient, and high-performing language models that can be used in a wide range of real-world applications.

2 Architectural details

Figure 1: Sliding Window Attention. The number of operations in vanilla attention is quadratic in the sequence length, and the memory increases linearly with the number of tokens. At inference time, this incurs higher latency and smaller throughput due to reduced cache availability. To alleviate this issue, we use sliding window attention: each token can attend to at most $W$ tokens from the previous layer (here, $W\mathrm{=}\mathrm{3}$). Note that tokens outside the sliding window still influence next word prediction. At each attention layer, information can move forward by $W$ tokens. Hence, after $k$ attention layers, information can move forward by up to $k\mathrm{\times }W$ tokens.

Parameter	Value
dim	$4096$
n_layers	$32$
head_dim	$128$
hidden_dim	$14336$
n_heads	$32$
n_kv_heads	$8$
window_size	$4096$
context_len	$8192$
vocab_size	$32000$

Table 1: Model architecture.

Mistral 7B is based on a transformer architecture vaswani2017attention . The main parameters of the architecture are summarized in Table 1. Compared to Llama, it introduces a few changes that we summarize below.

Sliding Window Attention. SWA exploits the stacked layers of a transformer to attend information beyond the window size $W$. The hidden state in position $i$ of the layer $k$, $h_i$, attends to all hidden states from the previous layer with positions between $i-W$ and $i$. Recursively, $h_i$ can access tokens from the input layer at a distance of up to $W\times k$ tokens, as illustrated in Figure 1. At the last layer, using a window size of $W=4096$, we have a theoretical attention span of approximately $131K$ tokens. In practice, for a sequence length of 16K and $W=4096$, changes made to FlashAttention dao2022flashattention and xFormers xFormers2022 yield a 2x speed improvement over a vanilla attention baseline.

Rolling Buffer Cache. A fixed attention span means that we can limit our cache size using a rolling buffer cache. The cache has a fixed size of $W$, and the keys and values for the timestep $i$ are stored in position $imodW$ of the cache. As a result, when the position $i$ is larger than $W$, past values in the cache are overwritten, and the size of the cache stops increasing. We provide an illustration in Figure 2 for $W=3$. On a sequence length of 32k tokens, this reduces the cache memory usage by 8x, without impacting the model quality.

Figure 2: Rolling buffer cache. The cache has a fixed size of $W\mathrm{=}\mathrm{4}$. Keys and values for position $i$ are stored in position $i\mathrm{mod}W$ of the cache. When the position $i$ is larger than $W$, past values in the cache are overwritten. The hidden state corresponding to the latest generated tokens are colored in orange.

Pre-fill and Chunking. When generating a sequence, we need to predict tokens one-by-one, as each token is conditioned on the previous ones. However, the prompt is known in advance, and we can pre-fill the ($k$, $v$) cache with the prompt. If the prompt is very large, we can chunk it into smaller pieces, and pre-fill the cache with each chunk. For this purpose, we can select the window size as our chunk size. For each chunk, we thus need to compute the attention over the cache and over the chunk. Figure 3 shows how the attention mask works over both the cache and the chunk.

Figure 3: Pre-fill and chunking. During pre-fill of the cache, long sequences are chunked to limit memory usage. We process a sequence in three chunks, “The cat sat on”, “the mat and saw”, “the dog go to”. The figure shows what happens for the third chunk (“the dog go to”): it attends itself using a causal mask (rightmost block), attends the cache using a sliding window (center block), and does not attend to past tokens as they are outside of the sliding window (left block).

3 Results

We compare Mistral 7B to Llama, and re-run all benchmarks with our own evaluation pipeline for fair comparison. We measure performance on a wide variety of tasks categorized as follow:

• Commonsense Reasoning (0-shot): Hellaswag zellers2019hellaswag , Winogrande sakaguchi2021winogrande , PIQA bisk2020piqa , SIQA sap2019socialiqa , OpenbookQA mihaylov2018can , ARC-Easy, ARC-Challenge clark2018think , CommonsenseQA talmor2018commonsenseqa

• World Knowledge (5-shot): NaturalQuestions kwiatkowski2019natural , TriviaQA joshi2017triviaqa

• Reading Comprehension (0-shot): BoolQ clark2019boolq , QuAC choi2018quac

• Math: GSM8K cobbe2021training (8-shot) with maj@8 and MATH hendrycks2021measuring (4-shot) with maj@4

• Code: Humaneval chen2021evaluating (0-shot) and MBPP austin2021program (3-shot)

• Popular aggregated results: MMLU hendrycks2020measuring (5-shot), BBH suzgun2022challenging (3-shot), and AGI Eval zhong2023agieval (3-5-shot, English multiple-choice questions only)

Detailed results for Mistral 7B, Llama 2 7B/13B, and Code-Llama 7B are reported in Table 2. Figure 4 compares the performance of Mistral 7B with Llama 2 7B/13B, and Llama 1 34B in different categories. Mistral 7B surpasses Llama 2 13B across all metrics, and outperforms Llama 1 34B on most benchmarks. In particular, Mistral 7B displays a superior performance in code, mathematics, and reasoning benchmarks.

Size and Efficiency. We computed “equivalent model sizes” of the Llama 2 family, aiming to understand Mistral 7B models’ efficiency in the cost-performance spectrum (see Figure 5). When evaluated on reasoning, comprehension, and STEM reasoning (specifically MMLU), Mistral 7B mirrored performance that one might expect from a Llama 2 model with more than 3x its size. On the Knowledge benchmarks, Mistral 7B’s performance achieves a lower compression rate of 1.9x, which is likely due to its limited parameter count that restricts the amount of knowledge it can store.

Evaluation Differences. On some benchmarks, there are some differences between our evaluation protocol and the one reported in the Llama 2 paper: 1) on MBPP, we use the hand-verified subset 2) on TriviaQA, we do not provide Wikipedia contexts.

Figure 4: Performance of Mistral 7B and different Llama models on a wide range of benchmarks. All models were re-evaluated on all metrics with our evaluation pipeline for accurate comparison. Mistral 7B significantly outperforms Llama 2 7B and Llama 2 13B on all benchmarks. It is also vastly superior to Llama 1 34B in mathematics, code generation, and reasoning benchmarks.

Model	Modality	MMLU	HellaSwag	WinoG	PIQA	Arc-e	Arc-c	NQ	TriviaQA	HumanEval	MBPP	MATH	GSM8K
LLaMA 2 7B	Pretrained	44.4%	77.1%	69.5%	77.9%	68.7%	43.2%	24.7%	63.8%	11.6%	26.1%	3.9%	16.0%
LLaMA 2 13B	Pretrained	55.6%	80.7%	72.9%	80.8%	75.2%	48.8%	29.0%	69.6%	18.9%	35.4%	6.0%	34.3%
Code-Llama 7B	Finetuned	36.9%	62.9%	62.3%	72.8%	59.4%	34.5%	11.0%	34.9%	31.1%	52.5%	5.2%	20.8%
Mistral 7B	Pretrained	60.1%	81.3%	75.3%	83.0%	80.0%	55.5%	28.8%	69.9%	30.5%	47.5%	13.1%	52.2%

Table 2: Comparison of Mistral 7B with Llama. Mistral 7B outperforms Llama 2 13B on all metrics, and approaches the code performance of Code-Llama 7B without sacrificing performance on non-code benchmarks.

Figure 5: Results on MMLU, commonsense reasoning, world knowledge and reading comprehension for Mistral 7B and Llama 2 (7B/13B/70B). Mistral 7B largely outperforms Llama 2 13B on all evaluations, except on knowledge benchmarks, where it is on par (this is likely due to its limited parameter count, which limits the amount of knowledge it can compress).

Model	Chatbot Arena ELO Rating	MT Bench
WizardLM 13B v1.2	1047	7.2
Mistral 7B Instruct	1031	6.84 +/- 0.07
Llama 2 13B Chat	1012	6.65
Vicuna 13B	1041	6.57
Llama 2 7B Chat	985	6.27
Vicuna 7B	997	6.17
Alpaca 13B	914	4.53

Table 3: Comparison of Chat models. Mistral 7B – Instruct outperforms all 7B models on MT-Bench, and is comparable to 13B – Chat models.

4 Instruction Finetuning

To evaluate the generalization capabilities of Mistral 7B, we fine-tuned it on instruction datasets publicly available on the Hugging Face repository. No proprietary data or training tricks were utilized: Mistral 7B – Instruct model is a simple and preliminary demonstration that the base model can easily be fine-tuned to achieve good performance. In Table 3, we observe that the resulting model, Mistral 7B – Instruct, exhibits superior performance compared to all 7B models on MT-Bench, and is comparable to 13B – Chat models. An independent human evaluation was conducted on https://llmboxing.com/leaderboard.

In this evaluation, participants were provided with a set of questions along with anonymous responses from two models and were asked to select their preferred response, as illustrated in Figure 6. As of October 6, 2023, the outputs generated by Mistral 7B were preferred 5020 times, compared to 4143 times for Llama 2 13B.

5 Adding guardrails for front-facing applications

The ability to enforce guardrails when it comes to AI generation is important for front-facing applications. In this section, we highlight how to leverage system prompting to optionally enforce output constraints on top of our models. Additionally, we showcase the ability of Mistral 7B to perform fine-grained content moderation, which can be useful to enforce quality content in applications.

5.1 System prompt to enforce guardrails

We introduce a system prompt (see below) to guide the model to generate answers within specified guardrails, similar to the work done with Llama 2. Using this prompt allows the user to move on the Pareto front of model utility / guardrails enforcement, as indicated in Table 4.

Always assist with care, respect, and truth. Respond with utmost utility yet securely. Avoid harmful, unethical, prejudiced, or negative content. Ensure replies promote fairness and positivity.

Guardrails	MT Bench
No system prompt	6.84 $\pm$ 0.07
Llama 2 system prompt	6.38 $\pm$ 0.07
Mistral system prompt	6.58 $\pm$ 0.05

Table 4: System prompts. Mean official MT Bench score over 10 iterations with standard deviation for Mistral 7B – Instruct. For reference, Llama 2 13B – Chat reports official results of 6.65.

We use a set of 175 unsafe prompts for evaluating safety. With the recommended system prompt, the model properly declines to answer 100% of the harmful questions.

As an illustration, we provide in Table 5 the answers of both Mistral 7B – Instruct and Llama 2 Chat 13B to the question How to kill a linux process with system prompts activated. We observe that Mistral 7B provides a correct response while Llama 2 declines to answer. Note that on this specific question, both models answer correctly when system prompts are deactivated.

Model	Answer
Mistral 7B – Instruct with Mistral system prompt	To kill a Linux process, you can use the `kill`command followed by the process ID (PID) of the process you want to terminate. For example, to kill process with PID 1234, you would run the command `kill 1234`. It’s important to note that killing a process can have unintended consequences, so it’s generally a good idea to only kill processes that you are certain you want to terminate. Additionally, it’s generally a good idea to use the `kill`command with caution, as it can be used to terminate critical system processes that are necessary for the operation of the system.
Llama 2 13B – Chat with Llama 2 system prompt	As a helpful and respectful assistant, I cannot provide information on how to kill a Linux process. It is not appropriate or ethical to intentionally harm or terminate processes, as it can cause damage to the system and potentially result in data loss. Additionally, it is important to respect the autonomy and security of other users’ processes, as they may be critical to their work or personal activities. Instead, I suggest that you explore other options to address any issues or concerns you may have.

Table 5: Comparison between Mistral and Llama system prompts. Mistral provides the right answer whereas Llama 2 declines to answer to the question How to kill a linux process.

5.2 Content moderation with self-reflection

Mistral 7B – Instruct can be used as a content moderator: the model itself is able to accurately classify a user prompt or its generated answer as being either acceptable or falling into one of the following categories: Illegal activities such as terrorism, child abuse or fraud; Hateful, harassing or violent content such as discrimination, self-harm or bullying; Unqualified advice for instance in legal, medical or financial domains.

To do so, we designed a self-reflection prompt that makes Mistral 7B classify a prompt or a generated answer. We evaluated self-reflection on our manually curated and balanced dataset of adversarial and standard prompts and got a precision of 99.4% for a recall of 95.6% (considering acceptable prompts as positives).

The use cases are vast, from moderating comments on social media or forums to brand monitoring on the internet. In particular, the end user is able to select afterwards which categories to effectively filter based on their particular use-case.

6 Conclusion

Our work on Mistral 7B demonstrates that language models may compress knowledge more than what was previously thought. This opens up interesting perspectives: the field has so far put the emphasis on scaling laws in 2 dimensions (directly associating model capabilities to training cost, as in hoffmann2022compute ); the problem is rather 3 dimensional (model capabilities, training cost, inference cost), and much remains to be explored to obtain the best performance with the smallest possible model.