stable-code

Please note: For commercial use, please refer to https://stability.ai/license.

Model Description

stable-code-3b is a 2.7B billion parameter decoder-only language model pre-trained on 1.3 trillion tokens of diverse textual and code datasets. stable-code-3b is trained on 18 programming languages (selected based on the 2023 StackOverflow Developer Survey) and demonstrates state-of-the-art performance (compared to models of similar size) on the MultiPL-E metrics across multiple programming languages tested using BigCode's Evaluation Harness.

Key Features

• Fill in Middle Capability (FIM)
• Supports Long Context, trained with Sequences upto 16,384

Usage

Get started generating text with stable-code-3b by using the following code snippet:

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("stabilityai/stable-code-3b")
model = AutoModelForCausalLM.from_pretrained(
  "stabilityai/stable-code-3b",
  torch_dtype="auto",
)
model.cuda()
inputs = tokenizer("import torch\nimport torch.nn as nn", return_tensors="pt").to(model.device)
tokens = model.generate(
  **inputs,
  max_new_tokens=48,
  temperature=0.2,
  do_sample=True,
)
print(tokenizer.decode(tokens[0], skip_special_tokens=True))

Run with Fill in Middle (FIM) ⚡️

from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("stabilityai/stable-code-3b")
model = AutoModelForCausalLM.from_pretrained(
  "stabilityai/stable-code-3b",
  torch_dtype="auto",
  attn_implementation="flash_attention_2",
)
model.cuda()
inputs = tokenizer("def fib(n):    else:\n        return fib(n - 2) + fib(n - 1)", return_tensors="pt").to(model.device)
tokens = model.generate(
  **inputs,
  max_new_tokens=48,
  temperature=0.2,
  do_sample=True,
)
print(tokenizer.decode(tokens[0], skip_special_tokens=True))

Run with Flash Attention 2 ⚡️

from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("stabilityai/stable-code-3b", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
  "stabilityai/stable-code-3b",
  trust_remote_code=True,
  torch_dtype="auto",
+ attn_implementation="flash_attention_2",
)
model.cuda()
inputs = tokenizer("import torch\nimport torch.nn as nn", return_tensors="pt").to(model.device)
tokens = model.generate(
  **inputs,
  max_new_tokens=48,
  temperature=0.2,
  do_sample=True,
)
print(tokenizer.decode(tokens[0], skip_special_tokens=True))

Model Details

• Developed by: Stability AI
• Model type: stable-code-3b models are auto-regressive language models based on the transformer decoder architecture.
• Language(s): English, Code
• Library: GPT-NeoX
• License: Stability AI Community License.
• Commercial License: to use this model commercially, please refer to https://stability.ai/license
• Contact: For questions and comments about the model, please email lm@stability.ai

Model Architecture

The model is a decoder-only transformer similar to the LLaMA (Touvron et al., 2023) architecture with the following modifications:

• Position Embeddings: Rotary Position Embeddings (Su et al., 2021) applied to the first 25% of head embedding dimensions for improved throughput following Black et al. (2022).
• Tokenizer: We use a modified version of the GPTNeoX Tokenizer.NeoX. We add special tokens to train for Fill in the Middle (FIM) capabilities like and along with other special tokens.

Training

Training Dataset

The dataset is comprised of a filtered mixture of open-source large-scale datasets available on the HuggingFace Hub: Falcon RefinedWeb extract (Penedo et al., 2023), along with CommitPackFT and Github Issues (BigCode., 2023), and StarCoder (Li et al., 2023). We further supplement our training with data from mathematical domains (Azerbayev, Zhangir, et al., 2023 and, Yu, Longhui, et al., 2023).

Top 18 programming languages trained on:

• C
• CPP
• Java
• JavaScript
• CSS
• Go
• HTML
• Ruby
• Rust
• Markdown
• Shell
• Php
• Sql
• R
• Typescript
• Python
• Jupyter-Clean
• RestructuredText

Training Procedure

The model is pre-trained on the aforementioned datasets in bfloat16 precision, optimized with AdamW.

Training Infrastructure

• Hardware: stable-code-3b was trained on the Stability AI cluster across 256 NVIDIA A100 40GB GPUs (AWS P